It's time to tell in detail how our implementation of the license plate recognition algorithm works: what turned out to be a good solution, what worked very badly. And just report to Habra-users - after all, with the help of the Recognitor Android application, you helped us gain a decent-sized database of pictures of license plates taken in a completely unbiased way, without explaining how to shoot and how not. And the database of images is the most important thing when developing recognition algorithms!

What happened with the Android application Recognitor

It was very nice that Habr users started downloading the application, trying it and sending us numbers.

Program downloads and ratings

Since the application was uploaded to the server, 3800 pictures of numbers from the mobile application have been received.
And even more we were pleased with the link http://212.116.121.70:10000/uploadimage - in 2 days we were sent about 8 thousand full-size pictures of license plates (mainly from Vologda)! The server was almost down.

Now we have a base of 12,000 photographs in our hands - there is a gigantic work ahead of debugging algorithms. All the fun is just beginning!

Let me remind you that a number was previously allocated in the Android application. In this article, I will not dwell on this stage in detail. In our case, the Haar cascade detector. This detector does not always work if the number in the frame is rotated too much. I will leave the analysis of how the trained cascade detector works when it does not work for the next articles. It's really very interesting. It seems that this is a black box - that's trained the detector and do nothing else. Actually it is not.

But still, a cascade detector is a good option in the case of limited computing resources. If the license plate is dirty or the frame is poorly visible, then Haar also performs well in relation to other methods.

Number recognition

Here is a story about text recognition in pictures of this kind:

General recognition approaches were described in the first article.

Initially, we set ourselves the task of recognizing dirty, partially erased and seriously distorted license plates.
Firstly, it's interesting, and secondly, it seemed that then pure ones would work in general in 100% of cases. Usually, of course, that's what happens. But it didn't work out here. It turned out that if for dirty numbers the probability of success was 88%, then for clean ones, for example, 90%. Although in reality the probability of recognition from a photo on a mobile application to a successful answer, of course, turned out to be even worse than the indicated figure. Slightly less than 50% of incoming images (so people don't try to take pictures). Those. on average, the license plate had to be photographed twice in order to recognize it successfully. Although in many respects such a low percentage is due to the fact that many tried to take numbers from the monitor screen, and not in real life.

The whole algorithm was built for dirty numbers. But it turned out that now in the summer in Moscow 9 out of 10 rooms are perfectly clean. So it is better to change the strategy and make two separate algorithms. If it was possible to quickly and reliably recognize a clean number, then we will send this result to the user, and if it was not possible, then we spend a little more processor time and run the second algorithm for dirty numbers.

A simple number plate recognition algorithm that should be implemented right away

How to recognize a good and clean number? It's not difficult at all.

We present the following requirements for such an algorithm:

1) some stability to turns (± 10 degrees)
2) resistance to minor scaling (20%)
3) cutting off any borders of the number by the border of the frame or simply poorly defined borders should not destroy everything (this is fundamentally important, because in the case of dirty numbers you have to rely on the border of the number; if the number is clean, then nothing better characterizes numbers / letters room).

So, in clean and well-read numbers, all numbers and letters are separable from each other, which means that you can binarize the image and use morphological methods to either select related areas, or use the well-known contour selection functions.

We binarize the frame

Here it is worth going through the mid-pass filter and normalizing the image.


The image shows an initially low-contrast frame for clarity.

Then binarize by a fixed threshold (you can fix the threshold, since the image has been normalized).

Frame Rotation Hypotheses

Let's assume several possible angles of rotation of the image. For example, +10, 0, -10 degrees:

In the future, the method will have little resistance to the angle of rotation of numbers and letters, so such a rather large step in the angle - 10 degrees was chosen.
We will work with each frame independently in the future. Which rotation hypothesis will give the best result will win.

And then collect all related areas. Here we used the standard function findContours from OpenCV. If the connected area (contour) has a height in pixels from H1 to H2, and the width and height are related by the ratio from K1 to K2, then leave it in the frame and note that there can be a sign in this area. Almost certainly at this stage only numbers and letters will remain, the rest of the garbage will leave the frame. Let's take the rectangles bounding the contours, bring them to the same scale, and then work with each letter / number separately.

Here are the bounding boxes of the paths that satisfied our requirements:

Letters/numbers

The image quality is good, all letters and numbers are perfectly separable, otherwise we would not have reached this step.
We scale all signs to the same size, for example, 20x30 pixels. Here they are:

By the way, when OpenCV performs Resize (when reduced to a size of 20x30), the binarized image will turn into a gradient one, due to interpolation. We'll have to repeat the binarization.

And now the easiest way to compare with known sign images is to use XOR (Normalized Hamming Distance). For example like this:

Distance = 1.0 - |Sample XOR Image|/|Sample|

If the distance is greater than the threshold, then we consider that we have found a sign, if it is less, we throw it out.

Letter-number-number-number-letter-letter

Yes, we are looking for Russian car signs in this format. Here you need to take into account that the number 0 and the letter "o" are not at all distinguishable from each other, the number 8 and the letter "c". We will line up all the signs from left to right and we will take 6 signs.
Criterion times - letter-number-number-number-letter-letter (don't forget about 0/o, 8/v)
Criterion two - deviation of the lower limit of 6 characters from the line

The total score for the hypothesis is the sum of the Hamming distances of all 6 signs. The bigger, the better.

So, if the total points are less than the threshold, then we consider that we have found 6 digits of the number (without the region). If it is more than the threshold, then we go to the algorithm that is resistant to dirty numbers.

Here it is still worth considering separately the letters "H" and "M". To do this, you need to make a separate classifier, for example, according to the histogram of gradients.

Region

The next two or three signs above the line drawn along the bottom of the 6 already found signs are the region. If the third digit exists and its similarity is greater than the threshold, then the region consists of three digits. Otherwise, out of two.

However, region recognition often does not go as smoothly as desired. The numbers in the regions are smaller, they may not be successfully divided. Therefore, the region is better recognized in a way that is more resistant to dirt/noise/overlapping, as described below.

Some details of the description of the algorithm are not disclosed in too much detail. Partly due to the fact that now only a mock-up of this algorithm has been made and it remains to be tested and debugged on those thousands of images. If the number is good and clean, then you need to recognize the number in tens of milliseconds or answer “failed” and move on to a more serious algorithm.

Dirty Number Resistant Algorithm

It is clear that the algorithm described above does not work at all if the characters on the license plate stick together due to poor image quality (dirt, poor resolution, bad shadow or shooting angle).

Here are examples of numbers when the first algorithm failed to do anything:

But you will have to rely on the boundaries of the license plate, and then, within a strictly defined area, look for signs with a precisely known orientation and scale. And most importantly - no binarization!

We are looking for the lower limit of the number

The simplest and most reliable step in this algorithm. We go through several hypotheses on the angle of rotation and build for each hypothesis on the rotation a histogram of the brightness of the pixels along the horizontal lines for the lower half of the image:

Let's choose the maximum of the gradient and thus determine the angle of inclination and at what level to cut off the number from below. Let's not forget to improve the contrast and get this image:

In general, it is worth using not only the brightness histogram, but also the dispersion histogram, the gradient histogram, to increase the reliability of the number clipping.

We are looking for the upper limit of the number

It’s not so obvious here, it turned out that if the rear license plate is removed from the hands, then the upper border can be strongly curved and partially cover the signs or in the shade, as in this case:


There is no sharp transition of brightness in the upper part of the number, and the maximum gradient will completely cut the number in the middle.

We got out of the situation in a not very trivial way: we trained the Haar cascade detector for each number and each letter, found all the signs in the image, and determined the top line where to cut:

It would seem that it is worth stopping here - we have already found numbers and letters! But in reality, of course, the Haar detector can be wrong, and we have 7-8 characters here. A good example of the number 4. If the upper border of the number merges with the number 4, then it is not at all difficult to see the number 7. Which by the way happened in this example. But on the other hand, despite the detection error, the upper border of the found rectangles really coincides with the upper border of the license plate.

Find the side borders of the number

Also nothing tricky - absolutely the same as the bottom one. The only difference is that often the brightness of the gradient of the first or last character in the number can exceed the brightness of the gradient of the vertical border of the number, so not the maximum is selected, but the first gradient that exceeds the threshold. Similarly, with the lower border, it is necessary to sort out several slope hypotheses, since due to the perspective, the perpendicularity of the vertical and horizontal borders is not at all guaranteed.

So here's a nicely trimmed number:

Yes! it is especially nice to insert a frame with a disgusting number that was successfully recognized.

Only one thing is sad - by this stage, from 5% to 15% of the numbers can be cut off incorrectly. For example, like this:

(by the way, someone sent us a yellow taxi number, as far as I understand - the format is not regular)

All this was necessary so that all this was done only to optimize calculations, since it is very computationally expensive to sort out all possible positions, scales and slopes of signs when searching for them.

Split string into characters

Unfortunately, due to the perspective and non-standard width of all signs, you have to somehow highlight the characters in the already cropped number. Here again, the histogram in brightness will help out, but along the X axis:

The only thing that is worth exploring in the future are two hypotheses: the symbols begin immediately or one histogram maximum should be skipped. This is due to the fact that on some numbers the hole for the screw or the head of the license plate screw may differ as a separate sign, or may be completely invisible.

Character recognition

The image is still not binarized, we will use all the information we have.

Here are printable characters, so a weighted covariance is suitable for comparing images with an example:

Samples to compare and weight under covariance:

Of course, you can't just compare the area highlighted with the horizontal histogram to the samples. We have to make several hypotheses on displacement and scale.
Number of hypotheses by position along the X axis = 4
Number of hypotheses by position along the Y axis = 4
Number of hypotheses by scale = 3

Thus, for each area, when compared with one sign, it is necessary to calculate 4x4x3 covariances.

First of all, we will find 3 large numbers. That's 3 x 10 x 4 x 4 x 3 = 1440 comparisons.

Then one letter on the left and two more on the right. There are 12 letters for comparison. Then the number of comparisons is 3x12x4x4x3 = 1728

When we have 6 characters, then everything to the right of them is the region.

There may be 2 digits or 3 digits in the region - this must be taken into account. Splitting the region in a histogram way is already meaningless due to the fact that the image quality may be too low. Therefore, we simply alternately find the numbers from left to right. Starting from the upper left corner, we need several hypotheses for the x-axis, y-axis and scale. Finding the best match. We shift by a given amount to the right, again we are looking for. We will search for the third character to the left of the first and to the right of the second, if the similarity measure of the third character is greater than the threshold, then we are lucky - the region number consists of three digits.

conclusions

The practice of applying the algorithm (the second one described in the article) once again confirmed the common truth in solving recognition problems: a truly representative base is needed when creating algorithms. We were aiming for dirty and shabby rooms, because the test base was filmed in winter. Indeed, it was often possible to recognize rather bad numbers, but there were almost no clean numbers in the training sample.

The other side of the coin was also revealed: few things irritate the user as much as the situation when the automatic system does not solve a completely primitive task. “Well, what can not be read here ?!” And the fact that the automatic system could not recognize a dirty or shabby number plate is expected.

Frankly speaking, this is our first experience in developing a recognition system for the mass consumer. And it is worth learning to think about such “little things” as about users. Now a specialist has joined us, who has developed a similar “Recognitor” program for iOs. In the UI, the user has the opportunity to see what is currently being sent to the server, to choose which of the numbers allocated by Haar is necessary, it is possible to select the required area in an already “frozen” frame. And it's more convenient to use it. Automatic recognition becomes not a stupid function without which nothing can be done, but simply an assistant.

Thinking through a system in which automatic image recognition would be harmonious and convenient for the user turned out to be no easier task than creating these recognition algorithms.

And, of course, I hope that the article will be useful.

Technologies for software recognition of car numbers and people's faces are becoming more and more in demand. For example, automatic license plate recognition can be used as a component of an access control system, to organize billing systems for paid parking, automate the passage of cars, or to collect statistical information (repeated visits to a mall or car wash, for example). All this is within the power of modern intellectual software. What is needed to implement such a system? In principle, not so much - video cameras that meet certain requirements and the corresponding intelligent software module. For example, software or more budget

In this article, we will tell you how to choose the right digital video camera capable of generating a high-quality video image that is acceptable for software license plate recognition tasks.

Permission

A few years ago, the size of the license plate on the screen was measured in % of the frame width. All television cameras were analog and their resolution was constant. Now, when matrices can have resolutions from 0.5 to 12Mp, relative values ​​do not apply and the required license plate width is measured in pixels.

As a rule, the specifications for license plate recognition software indicate the requirements for the width of the license plate on the screen, sufficient for their confident recognition. So, for example, the Autotrassir software module requires a width of 120 pixels, and NumberOK requires 80 pixels. Differences in the requirements are explained both by the nuances of the operation of recognition algorithms and by the acceptable level of reliability adopted by the developer. From personal experience, it can be noted that Autotrassir is more demanding and “capricious” in terms of choosing equipment, lenses, and the correct installation of the camera. But, being brought to mind, it shows consistently reliable results and depends little on weather conditions.

For greater reliability, we can recommend focusing on the value of the license plate width of 150 pixels. And if we remember that the width of the license plate according to GOST is half a meter (520mm to be precise), then we come to the required resolution of 300 dots per meter.

The linear resolution of pixels per meter depends on the viewing angle and the resolution of the camera matrix. It can be calculated using the formula:

Rlin- linear resolution, pixels per meter

R h- the horizontal resolution of the camera (for example,R h =1080)

𝛼 - camera angle

L- distance from the camera to the object

You can also use our online calculator on the page of the product you are interested in, on the “What I see” tab.

Below are (for example) several options for IP video surveillance cameras indicating the maximum distance from which license plate recognition is possible (license plate width 150 pixels). Please note that for cameras with a varifocal lens, the maximum focal length was used in the calculation

	Focal length	Horizontal Resolution	Max. distance, m	Max. viewing width, m
		1920 pixels
		1280 pixels
		2688 pixels
		2048 pixels
		2048 pixels

It is important to understand that cameras with higher resolution can cover wider areas, so fewer cameras are needed for the same area. In this case, the linear resolution remains within the limits of identification requirements. This fact makes it economically feasible to use high resolution cameras in many situations.

Light sensitivity and shutter speed

For reliable recognition of car license plates, the camera must have good light sensitivity and the ability to manually set the shutter speed (shutter speed or just shutter speed). This requirement is extremely important when building license plate recognition systems for vehicles moving at high speed. For cars moving at speeds up to 30 km/h (namely, we usually implement such projects for our customers: cottage settlements, residential complexes, shopping center parking lots, various closed areas) this requirement is less important, but it cannot be underestimated, because to achieve high recognition quality, the camera must take at least ten frames with a readable number.
Therefore, for example, to recognize the number plate of a car moving at a speed of 30 km/h with a camera installation angle of up to 10 degrees relative to the axis of motion, the shutter speed should be about 1/200 second. For many inexpensive cameras, even during the daytime in cloudy weather, such an exposure may not be sufficient, and the picture will turn out to be dark and / or noisy. Therefore, it is worth paying close attention to the size of the matrix and its quality. Ideally, use a specialized black and white camera with a CCD matrix. However, their price is very high and the resolution is usually no more than 1MP, which imposes serious restrictions on their applicability.
In general, you should not chase high resolution unless there are objective reasons for it. Relatively inexpensive ultra-high resolution cameras (4Mp, 5Mp and higher) are built on 1/3, 1/2.8 and, less often, 1/2.5-inch matrices. Cameras with a resolution of 1.3 and 2 megapixels have the same matrix size. As a result, the size of each photosensitive element in the 1.3MP camera is noticeably larger than in the 5MP camera, and the larger the size, the more light each photosensitive element can collect. That is why the IP cameras recommended by us for number recognition tasks rarely have a resolution of more than 2 megapixels.

Wide dynamic range (WDR), backlight compensation

The dynamic range of a camera determines the ratio between the maximum and minimum light intensity that its sensor can normally capture. In other words, this is the ability of the camera to transmit both brightly lit and dark areas of the image without distortion and loss. This parameter is very important for automatic license plate recognition, because helps to deal with the illumination of the camera by headlights. However, even the most advanced cameras with 140dB WDR are not always able to handle high contrast lighting. In this case, additional illumination of visible light or operating in the IR range is installed, highlighting the area in which the number plate is recognized.

Depth of field

Depth of field, or, in full, the depth of field of the depicted space (DOF) is the range of distances at which objects are perceived as sharp.

This setting is determined by focal length, aperture, and subject distance. The greater the depth of field, the larger the focus area and the more opportunities to “catch” a sufficient number of clear frames of a moving car.

Perhaps the maximum effect on depth of field is the lens aperture. The smaller the aperture, the greater the depth of field; the larger, the less depth of field. All of our recommended number plate recognition cameras are able to adapt to changing lighting conditions by automatically changing the aperture. It is recommended to adjust the focus of such cameras at the maximum aperture, when the depth of field is minimal.

The greater the distance from the camera to the object, the greater the depth of field, so do not try to place the camera as close to the recognition zone as possible. On the other hand, the longer the focal length, the smaller the depth of field. According to our practice, the optimal distance from the camera to am is in the range from 6 to 10 meters. Although it is not impossible and recognition from a distance of 100 meters.

Distortion

Many lenses distort the image slightly. The most common is the so-called "barrel" distortion of the picture. This is due to the magnification being larger in the center and smaller at the edges, resulting in a resizing of the object. So, if the same object falls into the center of the image and onto its edge, its dimensions at the edge will appear smaller. This may affect the possibility of identification.

The shorter the focal length, the stronger the distortion can be noticeable. Therefore, it is undesirable to use cameras with wide-angle lenses (less than 4 mm) for identification.

Noise and color reproduction

The less noise and the more accurate the color reproduction, the better for identification. Therefore, it is recommended to pay attention to such parameters as the minimum illumination of the camera, as well as the presence of noise reduction functions.
Noise suppression is especially important in low light, when the camera sensors are very “noisy”, which complicates identification. It should be understood that in many cases noise reduction and other electronic “gadgets” cannot cope, and it is necessary to ensure a sufficient level of lighting at the facility.

Video compression

Modern IP cameras transmit a compressed video signal, and if there is no movement in the frame or it is minimal, the traffic will be small. If the movement in the frame is intense, the traffic will grow. Therefore, if a constant bitrate is set in the camera settings, the picture will be suitable for identification in the absence of movement, but unusable - with heavy movement in the frame.
For identification, it is recommended to set the variable bitrate with the highest quality level. In this case, the desired image quality will be provided.

Sensor: 1/2.8” Progressive Scan CMOS Hardware WDR 140dB Lens: 2.8-12mm Features: the chamber is internal, a thermal casing is required for installation on the street. Lens not included and sold separately
Max. Resolution: 1.3MP, 1280 x 960 pix Hardware WDR Lens: 2.8-12mm
AXIS P1365-E 2 MP Outdoor Network Camera with WDR and Lightfinder Sensor: 1/2.8” Progressive Scan CMOS Max. resolution: 2mp, 1920 x 1080 pix Hardware WDR Lightfinder Technology Lens: 2.8-8mm @F1.3 Features: High sensitivity, auto focus
Dahua IPC-HF8301E Utlra WDR 120dB, Ultra 3DNR Sensor: 1/3" Progressive Scan CMOS Max. resolution: 3mp, 2048x1536 pixels Hardware WDR Lens: 2.8-12mm Features: the chamber is internal, a thermal casing is required for installation on the street. Lens not included and sold separately
Sensor: 1/3” Progressive Scan CMOS Max. Resolution: 1.3MP, 1280x960 pix Lens: 2.8 - 8mm (F1.2) Features: High sensitivity, auto focus

1.1 Cameras

DS-2CD4A25FWD-IZ(H)(S) Lightfighter bullet and

DS-2CD4A26FWD-IZ(H)(S) Darkfighter bullet

Outdoor IR Bullet Camera

1. Works in very low light
2. excellent work of headlight compensation,
3. cylindrical, all-weather, rugged housing,
4. license plate recognition license,
5. b/w filter list,
6. alarm output

1. Darkfighter Ultra-low light technology High definition 1920x1080
2. Up to 60fps at Full HD1080p 120dB WDR
3. 2.8~12mm motorized VF lens with intelligent autofocus
4. Smart codec H.264+ compact compression IR illuminator 50m.
5. IP67 protection
6. Power supply +-12V DC and PoE
7. Built-in storage, support up to 128GB
8. Support ANPR, B/W List Filtering

Options:

Integrated heater (-H)

Audio/Alarm Inputs/Outputs (-S)

Cameras suitable for license plate recognition are initially supplied with firmware for counting passers-by,

therefore, at the request of the customer, they are reflashed for this function.

Flashing does not completely remove the counting function and allows you to return to it if desired by selecting the SMART Event, as shown in the figure below.

1.2 Solution

Solutions Hikvision license plate recognition provided by the camera itself can be divided into:

one). Classic number plate recognition and output of a list of recognized ones directly from the camera

1. Actions when the number matches the list recorded in the camera (admission to the territory, turning on the siren, sending a message)

When a number appears from the list, the dry contact of the camera is closed, which is a signal for the barrier control unit.

2. Camera requirement and installation location

2.1. The license plate must be legible and well lit.

2.2. The license plate must be at least 150 pixels wide.

2.3. Permissible tilt - no more than 5 ° (clockwise and counterclockwise).

2.4. Vertical angle - no more than 30 °.

The original formula is �=ℎ∗√3.

2.5. Horizontal angle - no more than 30 °.

2.6. If it is necessary to recognize license plates from two lanes, as a rule, it is recommended to place the camera on the crossbar.

2.7. It is necessary to choose the correct distance from the camera to the place of recognition

2.8. When recognizing license plates at night, IR illumination is required.

2.9. The shutter speed should be fast enough to reduce headlight glare at night. As a rule, we are talking about 1⁄1000.

2.10. Depth of focus is a very important parameter. If you are using a camera with a CS mount lens, use a fixed lens. Fixed lenses are better for recognition due to the greater depth of focus.

2.11. When choosing a mounting location, remember that direct sunlight can distort the picture.

2.12. When installing the camera on the side of the road, check how the support reacts to the passage of heavy vehicles or convoys of vehicles. If the support has tangible fluctuations, this will affect the efficiency of the system.

2.14. In rare cases, a situation of false detections may occur.
To minimize this, you need the following:

1. select the recognition zone correctly.
2. try changing the angle of view or the location of the camera.
3. adjust the settings for the minimum and maximum license plate size in the settings.

3. Vehicle detection

Intelligent IP camera detects the car by identifying and recognizing the license plate, giving the following data to the registrar, iVMS-5200 or other consumer:

1. Travel time (hours and minutes)
2. Direction of travel ("entry" and "exit" when choosing a travel zone)
3. License plate (letters and numbers)
4. Country of registration (name)
5. Screenshot with number (small picture)
6. Full screen screenshot
7. Video of the moment of determination (+/- 1-5s)
8. Auto-recognition of black / white lists (issuance of the corresponding alarm)
9. Activation of the alarm output relay (on the camera itself, in the registrar it is configured separately)

The information received from the camera is managed by the respective consumers:

You can set up the transfer of information and the actual recognition of license plates on the camera on the following consumers:

a) Recognition setting on localNVR

If the NVR is connected to the iVMS-4200, then the recorder and camera can be configured from it:

b) Recognition viVMS-4200

And even in the iVMS-4200, you can perform all the management of the recognition process, but independently without the NVR, it is just a shell that can only use the normal video surveillance functions from these cameras.

c) Recognition setting on theiVMS-5200 P

iVMS-5200 Pro has advanced analytics that uses number recognition in various activities of society and business.

Recognition setting on camera

On the camera itself, via the web-inteface, you can configure for any consumer, adjusting it already on it, but to connect the actuator, the configuration is done only on the camera.

Just here we will consider the function of detecting a car to open a barrier.

4. Camera setup

4.1. To process the license plate recognition event, such as, for example, the opening of a barrier, first of all, you need to configure the "Alarm output", by closing the dry contact of which the mechanism will work.

Without this, there will be no electrical reaction to the recognition that can be adjusted further.

However, if you do not plan to use mechanization, then this is not necessary.

4.2. As a rule, it is not supposed to open the barrier to all visitors, but only to “their own”, or, in extreme cases, not to let in only certain ones. Therefore, it is necessary to enter ahead of time the "white" and "black" list of numbers, for which you need to get its form from the camera itself by pressing the "export" button.

I want to pay attention to the name of the font used in the document, which, of course, is not in your system, but it is necessary for the camera to correctly perceive your presets:

After filling out the file with the list of numbers, you need to select the filled cells of the title bar, making sure that the font is named in Chinese, and then use the Excel button to copy the format according to the sample

Then you need to apply this format to all the cells you entered, highlighting them so that they are all written in a font with a Chinese name.

4.3. After importing the prepared Excel file, the data of the "white" and "black" lists will be filled in the camera:

Note: Unfortunately, with lists bye somewhat sad:

And now, only after everything previously done, you can start setting up license plate recognition and enabling the trigger reaction to numbers from the "white" list

4.4. Set the number of recognition strips and set up the zone, and then select the region.

Supported countries in the "EU and CIS" option:

Czech Republic, Germany, Spain, France, Italy, Netherlands, Poland, Slovakia, Belarus, Moldova, Ukraine, Russia, Belgium, Bulgaria, Denmark, Finland, Great Britain, Greece, Croatia, Hungary, Israel, Luxembourg, Macedonia, Norway, Portugal, Romania, Serbia, Azerbaijan, Georgia, Kazakhstan, Lithuania, Turkmenistan, Uzbekistan, Latvia, Estonia, Austria, Albania, Bosnia and Herzegovina, Republic of Ireland, Republic of Iceland, Vatican City, Republic of Malta, Sweden, Switzerland, Cyprus, Turkey, Slovenia.

4.5. Select the Entry/Exit mode.

4.6. Check and resave the schedule.

4.7. Turn off "All" by selecting "White List" and turn on the alarm output.

4.8. Enable recognition and save settings.

5. Number recognition

1. The recognition process can be seen in a special settings tab.

1. However, the results of detection and recognition can only be seen in the registrar's archive.
2. The camera will be able to write to its own memory card only constantly and on events:

License plate identification screenshots can also be sent to an FTP server by checking the Communication Method section of the Configuration Search tab of the Traffic menu.

6. Conclusion

Dont be upset! NVR, iVMS-4200 & 5200 do not have all the above problems! There everything works correctly and has great functionality!

A couple of months ago, Hikvision officially introduced firmware for cameras of the 4th series with the ability to recognize license plates. This firmware can be downloaded from the official website of Hikvision in Russia, the firmware is absolutely free and is suitable even for cameras purchased more than 2 years ago, the only drawback at the moment is the disabling of all smart functions, except for number recognition (this is due to the high load to the camera processor). At the moment, any owner of Hikvision 4 series cameras can test the possibility of this firmware on their cameras. The platform built into the camera detects and recognizes license plates, and sends the received information to or for access control.

For the correct performance of the license plate recognition function in the Hikvision 4th series camera, a number of parameters must be observed:

• for recognition, the horizontal angle is critical, and it should be within 0-7 degrees;
  
• the number must occupy at least 130 pixels in the image;
• camera firmware specialized 5.3.0_150719. So far only in English

What do we get via browser:

For private purposes, the license plate picture and one character from the recognized number were removed.

As you can see, the camera recognizes the license plate, indicates the time, region and photos of the number itself, if the number is not recognized for some reason, a screenshot is taken and it can be recognized manually. To save data in case of loss of connection, there is a buffer in the camera for 1000 numbers.

What do we get via Smart NVR(in our case, with firmware V3.4.0):

We receive a set of photos with the date, camera number and recognized vehicle number. The Registrar collects and stores this information. It is possible to search both by time and by number, by any number and letter, for any period of time. All the necessary information can be downloaded in an XLS file.

And also, you can upload back the file with numbers from the black and white list, and, thereby, automate the opening of the actuator.

The last option for obtaining information is. Today, it is this system that fully supports all the functionality of license plate recognition and other intelligent camera functions. In iVMS-5200Pro, the License Plate Recognition interface displays the last 8 license plates in real time to the right of the Smart Camera video with the time.

For private purposes, one character was removed from the recognized number.

At the same time, by clicking on the license plate, a detailed picture appears:
Search for information: all numbers and images of cars fall into the Database, where they can be easily searched for by parameters ( can be by any number and letter of the number, for any period). To do this, open the Number Request interface and indicate the information we are interested in.

We tested this system on 2 cameras + DS-7616NI-E2 with firmware V3.4.0 and iVMS-5200Pro software from Hikvision (not so long ago, this module also appeared in the updated version IVMS-4200 2.3.1.3 and was tested). Percentage of recognized license plates was ~70% , but, in our case, this was due to the impossibility of placing the camera perfectly and observing all the parameters for the angle of the horizon to the car number.

From the experience of our colleagues from St. Petersburg: “According to the information from the manufacturer, the percentage of recognized license plates is 85% at a speed of 65 km/h. In practice, sometimes 100+ cars drive along the embankment, but the numbers are recognized correctly. As for license plate recognition in the dark, according to information from the manufacturer, and, based on common sense, it is necessary to install an IR illuminator with a wavelength of 850 nm. In our case, the camera is mounted high and the spotlight is not suitable.”

In the above, we used a "bunch":

2) Tamron 5-50 lens

3) Thermal housing Hikvision 1313HZ-S

In general, the system works decently, license plate recognition occurs directly on the camera, the load on the client's PC is minimal (even if all cameras work with recognition), the system can read information from 4 lanes simultaneously.

The economic part of this solution: Hikvision 4th series cameras cost from 19990 rubles. separately you need a thermal casing and a good lens), and 39,990 rubles. for if you add a 16-channel recorder here, this is an additional 16,990 rubles. The cost of a license plate recognition system with one camera (an option for 39,990 rubles) based on a DVR will cost 57,000 rubles. When using 2-3 cameras, the price of the solution per channel will decrease and actually equal the cost of the camera itself.

Official dealer of Hikvision (Hikvision) in the Urals (C) 2017

There are many systems for automating the entry of cars into the territory of a protected facility. Starting from a banal guard in a booth with a button and ending with an electronic pass or a radio key fob.

The electronic license plate recognition system stands alone in this list and has not been very popular until recently.

There are several reasons for this.

Firstly, the high cost of equipment and the complexity of customization. Secondly, the active rejection of innovation, including acts of undisguised sabotage, by the guards themselves, whose work is now tightly controlled, excluding the possibility of additional earnings.

However, there are significant advantages that the license plate recognition system provides:

• a significant increase in the level of security and control of road transport at the facility;
• the possibility for third parties to enter the protected area using fake or stolen magnetic passes or electronic key fobs is excluded. (a car can also be stolen, but it is much more difficult);
• automatic reporting of vehicles with the ability to generate multiple reports;
• remote access capabilities allow the management of the organization to control the work of employees;
• the license plate recognition system can be easily integrated into the overall access control system of the organization.

The possibility of entering the territory of a protected facility by gluing the numbers printed on the printer to the car number is completely excluded. Virtually all license plate recognition systems control light reflectance, which paper does not have. The re-glued number simply will not be read.

The scope of automated license plate recognition systems is quite diverse. First of all, car number recognition will be useful at service stations, gas stations, car washes, warehouses, enterprises, parking lots.

The functions that such an automatic license plate recognition system can perform are quite diverse:

• control of entry and exit to the controlled territory;
• restriction of departure from the territory of the enterprise, for example, a bus station, a client who has not made a payment;
• monitoring the loading of the service area.

When combined with access control systems, license plate identification provides additional benefits. First of all, this is a complete control of the location of vehicles in the loading area of ​​the enterprise. This makes it possible to track the import of raw materials or the export of finished products, check the efficiency of loading and unloading operations and prevent theft.

At the same time, checking the car number not only at the entrance, but also at the exit excludes the possibility of exporting the goods using forged or erroneous accompanying documents.

But the owner of a parking lot or car park receives the most benefits. The automatic number plate recognition system will allow monitoring the occupancy of the territory in real time, which will make it possible to take measures to improve efficiency.

Combining license plate recognition with a payment system will completely eliminate the possibility of abuse or theft by employees. And it will also completely eliminate the possibility of errors in calculating the time spent by the vehicle in the parking lot and will give iron proof in disputes with unscrupulous customers.

TECHNICAL CHARACTERISTICS AND COMPOSITION OF EQUIPMENT

The system for automatic license plate recognition, depending on the manufacturer and model, may include several devices and a software package with modules that perform various analytics functions or serve atypical devices. For example, truck scales, speed radar, etc.

Requirements for the computer on which the program will be installed.

The minimum requirements for different programs can vary significantly depending on the functional load, but in most cases it is necessary:

• processor, at least 3 GHz;
• video card: Intel, ATI with OpenGL or nVidia at least 512 MB;
• RAM, not less than 4 GB;
• HDD disk with a capacity of at least 4 GB.

DVR with RTSP function.

This is a streaming protocol that allows not only viewing and recording information, but also using video in real time. An example of such recorders is the HIKVISION DS-7204HVI-SV model.

Surveillance camera with RTSP function.

Such devices for recognizing a car number must have a resolution of at least 550 TVL, which is provided by a 1/3 "760H matrix. The focal length is 9-22 mm, which will make it possible to identify at a considerable distance and at a fairly high speed, for example, Atis AW-CAR40VF or AW-CAR180VF.

The light sensitivity of the camera should be as high as possible from 0.001 Lux, in addition, the device must be equipped with IR illumination, which allows high-quality shooting from a distance of at least 15-20 m. The following functions are required:

• manual exposure setting;
• automatic white balance;
• backlight compensation;
• extended dynamic range.

These cameras will be used exclusively outdoors, so it is mandatory to have an IP 66 housing protection class with built-in thermocouples that allow the device to operate at low temperatures of at least -30°C.

It is recommended to use black and white cameras as they have higher sensitivity and resolution than color cameras. In addition, most license plate recognition algorithms convert the color image received from the camera to black and white.

Executive devices and control modules.

For example, the BARBOS module connected to a PC via a USB connection. This module has 4 five-ampere relays through which you can control the barrier, gate, gate, lighting, GSM notification, various indication systems displayed in the control room, etc.

CAMERAS FOR PLATE RECOGNIZATION

The main parameter that you should pay attention to when choosing a place to install CCTV cameras for license plate recognition is the manual setting of the shutter speed. There is a linear relationship between vehicle speed and the recommended shutter speed (frame exposure time - shutter).

The higher the speed of the car, the shorter the exposure time should be, otherwise the frame will be blurred - motion blur. However, the maximum allowable shutter speed depends not only on the exposure time, but also on the angle of the camera. The camera installation angle is the angle between the vehicle's direction of travel and the camera's optical axis.

Most mid-range cameras are capable of transmitting an 80-pixel wide license plate image suitable for recognition at a vertical installation angle of up to +30° and horizontal deflection angles of +/- 30°. It is considered a good indicator if the system recognized the license plate when it deviated from the horizontal (roughness of the road) +/- 10°.

A graph of the dependence of the exposure time on the camera installation angle and vehicle speed is shown in the figure.

Software.

Software is a key element of the license plate recognition system. There are many development companies offering their product to the consumer.

The most common budget development "NumberOK".

It recognizes Russian, Ukrainian, Belorussian and Moldavian license plates, records the date and time of entry and exit of vehicles and the time spent on the territory of the facility. It has the ability to build simple reports and can be integrated into 1C. The program is compatible with most camcorders and DVRs that have the RTSP function.

The second most important is the license plate recognition system. "Automarshal".

It has 2 recognition algorithms, one for speeds up to 30 km / h, the second - up to 150 km / h. It has specially adapted modules "Parking", "Car wash", "Gate ACS". Extensive opportunities for building analytical reports, management through the WEB client and the function of sending SMS notifications.

The number plate identification system has more extensive additional features. "Traffic control" research and production association "Diskret".

This program can connect to truck scales and link gross and net values ​​to the number, as well as generate summaries, balances and other reporting documents. "Traffic Control" maintains a photo archive of the moments of vehicles passing through the checkpoint and has ample opportunities for analytical search, by car or camera number, time and date.

System "Auto number" from the company "ELVIS Neo Tech".

The structure includes modules "Auto-control", "Senesys-Avto" and "Auto Number". The program has significant integration with other video surveillance systems and access control systems, as well as a flexible report generator, good archiving and search capabilities.

Undoubtedly, professional license plate recognition systems are quite expensive. And the use of an adapted conventional video surveillance system and demo versions of specialized software is not as effective as we would like.

But the use of this kind of video analytics can bring the business associated with road transport to a qualitatively new level, both in terms of control and business analysis.

* * *

© 2014-2020 All rights reserved.
The site materials are for informational purposes only and cannot be used as guidelines and normative documents.