CN105200938B - A Vision-Based Anti-Collision System for Lane Brake Lever - Google Patents

Abstract

The invention relates to the field of lane brake lever control systems. A vision-based lane gate anti-collision system, including a barrier gate, a processing module, an image acquisition module, a memory module, and a host computer; the image acquisition module collects images of target areas, and the processing module generates a hash of the corresponding image through a hash algorithm. The Greek fingerprint, by comparing the hash fingerprint to judge the template image and the real-time image at the beginning, to obtain whether there is a car in the real-time image, to determine whether the barrier of the barrier is down. The hash algorithm of the present invention processes images, and has the advantages of good robustness, good real-time performance, high stability, low algorithm complexity, and little influence of light changes; Prevent the brake lever from colliding with the vehicle.

Description

A Vision-Based Anti-Collision System for Lane Brake Lever

technical field

The invention relates to the field of lane brake lever control systems, in particular to a vision-based lane brake lever anti-collision system.

Background technique

Expressway toll stations are an important application field of intelligent transportation systems. In order to make the functions of cars and roads intelligent, improve transportation efficiency, ensure traffic safety, alleviate traffic congestion, and reduce environmental pollution, real-time, accurate, and efficient anti-smashing systems are particularly important. important. Due to the different chassis heights and metal contents of different vehicles, it is difficult to determine the sensitivity range of the traditional ground induction coil, resulting in failures such as the pole falling before the vehicle passes or the pole not falling after the vehicle passes. In addition, the ground induction coil will be deformed under the rolling of a large vehicle, and even cause irreversible damage, resulting in a decrease in the stability of the induction coil.

Contents of the invention

The purpose of the invention is: to solve the above technical problems, provide a vision-based lane brake lever anti-collision system, make up for the shortcomings of the ground induction coil, and propose an anti-smashing system solution based on image recognition, which has high stability , good real-time performance, strong anti-interference ability, and has great application value.

The technical solution of the present invention is: a vision-based lane barrier anti-collision system, including a barrier gate, a processing module, an image acquisition module, a memory module and a host computer; the image acquisition module is installed next to the barrier gate and faces the target area; the processing module analyzes the image information collected by the image acquisition module, and controls the state of the barrier of the barrier; the memory module stores a template image, and the processing module receives the triggering of the barrier from the upper computer of the lane When the pole signal is used, the control memory module updates the image information collected by the image acquisition module into a template image; the processing module performs Gaussian filtering on the image information to obtain an instant image; the processing module generates the hash of the instant image and the template image through a hash algorithm Fingerprint, hash fingerprint is an image description string combined in the same order; when the Hamming distance of the hash fingerprint between the instant image and the template image is greater than the set value, the gate will keep the railing erect; When the distance is less than the set value, the barrier of the gate will be lowered.

The image-based processing and analysis method of the present invention, specifically, the image acquisition module collects the image of the target area, the processing module generates the hash fingerprint of the corresponding image through the hash algorithm, and judges the template image and the instant image at the beginning by comparing the hash fingerprint, Get instant images of vehicles to determine if barriers are down.

Preferably, the processing module generates the hash fingerprint of the instant image and the hash fingerprint of the template image through the DCT hash algorithm, and calculates the Hamming distance M between the two; the processing module generates the hash of the instant image through the mean value hash algorithm. The hash fingerprint of the fingerprint and the template image is calculated, and the Hamming distance N between the two is calculated; when M and N are both greater than the set value, the gate will keep the valve stem upright; when M and N are less than the set value, the gate valve will The rod is lowered.

Here, two hash algorithms are used to analyze the image at the same time, which can effectively avoid the incompleteness of the algorithm, enhance the anti-interference ability of the algorithm, and then improve the reliability and stability of the algorithm.

Preferably, the processing module generates the 64-bit hash fingerprint of the instant image and the 64-bit hash fingerprint of the template image through the DCT hash algorithm, and calculates the Hamming distance M between the two; the processing module generates through the mean value hash algorithm Calculate the Hamming distance N between the 64-bit hash fingerprint of the real-time image and the 64-bit hash fingerprint of the template image; if M>10 and N>20, keep the railings upright; if M<5 and N<10 , The railing of the barrier gate was lowered.

Preferably, the hash algorithm is a DCT hash algorithm, and the hash fingerprint generation step of the instant image or template image is:

1) Reduce the image to a size of 32×32,

2) convert the reduced image into a grayscale image,

3) performing DCT transformation on the grayscale image, and obtaining a 32×32 DCT coefficient matrix,

4) retaining the 8×8 matrix in the upper left corner of the DCT coefficient matrix,

5) calculation step 4) coefficient mean value in the DCT coefficient matrix,

6) Compare the coefficients in the DCT matrix in step 4) with the average value one by one, if it is greater than or equal to the average value, record it as 1; if it is less than the average value, record it as 0; combine the comparison results in order to form 64 Bit hash fingerprint.

Here, the DCT hash algorithm or the DCT perceptual hash algorithm can quickly identify the difference between two images, specifically the difference between the "car" image and the "no car" image, so as to identify whether there is a car in the image. exist.

Preferably, when the Hamming distances of the hash fingerprints of both the instant image and the template image are greater than 10, the barrier of the barrier is kept erect; when the Hamming distance between the two is less than 5, the barrier of the barrier is lowered.

Preferably, the hash algorithm is a mean value hash algorithm, and the hash fingerprint generation step of the instant image or template image is:

1) Reduce the image to a size of 8×8, a total of 64 pixels;

2) Convert the 8×8 image into a grayscale image, and convert it into a 64-level grayscale;

3) Calculation step 2) grayscale mean of all 64 pixels;

4) If it is greater than or equal to the average value, record it as 1; if it is less than the average value, record it as 0; combine the comparison results in order to form a 64-bit hash fingerprint.

Here, the mean value hashing algorithm or the mean value perceptual hashing algorithm is simple and fast, and can quickly distinguish the difference between two images under the same background, specifically, distinguishing the difference between the "car" image and the "no car" image, so that Identify whether there is a car in the image.

Preferably, when the Hamming distances of the hash fingerprints of both the instant image and the template image are greater than 20, the barrier of the barrier remains erect; when the Hamming distances of both are less than 10, the barrier of the barrier is lowered.

Preferably, the image acquisition module includes a sequentially connected CCD image sensor and a SAA7113 video decoding chip; the processing module adopts a DM642 chip; the SAA7113 video decoding chip is connected to the VP1 interface of the DM642 chip, and the DM642 chip is connected via I ² C The bus connection controls the SAA7113 video decoder chip.

The beneficial effects of the present invention are:

The invention processes the image through the hash algorithm, which has the advantages of good robustness, good real-time performance, high stability, low algorithm complexity and little influence of illumination changes; it can effectively judge whether there are cars or no cars in the target area of the lane, Effectively prevent the brake lever from colliding with the vehicle.

The invention can effectively avoid the defect of the ground induction coil in the application of road induction, and provides an anti-collision system of the lane brake bar with high induction efficiency, accuracy and durability.

Description of drawings

Fig. 1 is a schematic diagram of functional modules of the present invention;

Fig. 2 is a schematic diagram of the workflow of the present invention;

Fig. 3 is DCT coefficient matrix;

Figure 4 is hash fingerprint example 1;

Figure 5 is hash fingerprint example 2.

Among them, 1—image acquisition module, 2—processing module, 3—memory module, 4—host computer, 5—gate, 6—power supply.

detailed description

The invention discloses a vision-based lane barrier anti-collision system, which includes a barrier, a processing module, an image acquisition module, a memory module and a host computer; the image acquisition module is installed beside the barrier and faces the target area; The processing module analyzes the image information collected by the image acquisition module, and controls the state of the barrier of the barrier; the memory module stores a template image, and the processing module receives the signal of the barrier from the upper computer of the driveway , the control memory module updates the image information collected by the image acquisition module into a template image; the processing module performs Gaussian filtering on the image information to obtain an instant image; the processing module generates hash fingerprints of the instant image and the template image through a hash algorithm, The hash fingerprint is an image description string combined in the same order; when the Hamming distance of the hash fingerprint between the instant image and the template image is greater than the set value, the gate will keep the railing up; when the Hamming distance between the two is less than When the value is set, the barrier of the barrier gate is lowered.

Here, the image is processed through the hash algorithm, and through the difference between the images, it can quickly and effectively identify whether there is a car in the image, so as to control the state of the barrier of the barrier and prevent the barrier from hitting the vehicle. Specifically, the image acquisition module collects the image of the target area, the processing module generates the hash fingerprint of the corresponding image through the hash algorithm, and judges the template image and the instant image at the beginning by comparing the hash fingerprint to obtain whether there is a car in the instant image. Determine if the gate valve stem is lowered. Compared with the existing anti-smashing system, the application of the induction coil is reduced, the structure is more stable and compact, and the application efficiency, accuracy and stability of the system are significantly improved compared with the existing anti-smashing system.

The implementation of the present invention will be described below in conjunction with the accompanying drawings.

As shown in FIG. 1 , it is a schematic diagram of functional modules of the present invention, including an image acquisition module 1 , a processing module 2 , a memory module 3 , a host computer 4 , a gate 5 and a power supply 6 .

The image collection module 1 is configured to collect image information of a target area, specifically, the target area is the lane area located in front of the railing in the driving direction. The image acquisition module 1 includes a CCD image sensor and a SAA7113 video decoding chip; the SAA7113 decodes the data signal collected by the CCD into a standard "VPO" digital signal and outputs it to the processing module 2, which is equivalent to an "A/D" device. Wherein, the processing module 2 is also connected to control the SAA7113 video decoding chip through an I ² C two-wire serial bus.

The processing module 2 is used to analyze the image signal, and is connected to control the state of the railing of the barrier gate 5. Processing module 2 adopts DM642 (full name TMS320DM642) in TI's C6000 series DSP. The core of DM642 is C6416 high-performance digital signal processor, which has strong processing performance, high flexibility and programmability, and at the same time integrates very Complete audio, video and network communication equipment and interfaces can meet the video/image processing in this system and complete the connection control of related equipment.

The memory module 3, specifically the peripheral memory module, is connected to the EMIF interface of the DM642 through an external bus. The memory module 3 includes SDRAM synchronous dynamic random access memory and FLASH solid state memory.

The upper computer 4 includes functions such as opening local pictures, setting IP addresses, receiving images, selecting target areas, sending area coordinates, and saving images.

Barrier gate 5 is used for channel entrance and exit management settings for restricting motor vehicles on the road. DM642 is connected to control the 5 railings of the barrier, specifically the motor connected to the control railing, to complete the state control of the 5 railings of the barrier.

The power supply 6 provides power supply for the system, specifically an independent power supply system can be used.

As shown in Figure 2, it is a schematic diagram of the workflow of the present invention. The specific working steps are as follows:

1. System initialization, including template image initialization in the storage module.

2. The processing module monitors whether the upper computer sends a signal to open the gate. If issued, go to the next step; if not, continue monitoring.

3. The processing module receives the image information sent by the image acquisition module, and updates the image information to the template image in the storage module, and the template image is locked.

4. Gaussian filtering is performed on the image information to obtain an instant image.

5. The processing module processes and analyzes the instant image, specifically based on the hash algorithm, including the DCT (perceptual) hash algorithm and the mean (perceptual) hash algorithm.

Among them, the DCT hash algorithm process includes:

1) Size reduction: The instant image is reduced to a size of 32×32, which simplifies the calculation of DCT.

2) Simplify color: convert the reduced real-time image into a grayscale image, further simplifying the amount of calculation.

3) Calculating DCT: Calculate the DCT transformation of the picture: F(u,v)=C ^T f C to obtain a 32×32 DCT coefficient matrix:

4) Reduce DCT: According to the DCT transformation formula F(u, v)=C ^T f C of 3), the DCT coefficient matrix with a size of 32×32 can be obtained, and the 8×8 matrix in the upper left corner of the DCT coefficient matrix is reserved, This part presents the lowest frequencies in the picture; see Figure 3.

5) Calculate the average value: calculate the average value of DCT, such as:

6) Generate a hash fingerprint: compare the 8×8 DCT coefficients with the average value, if it is greater than or equal to the average value, it will be recorded as 1, and if it is less than the average value, it will be recorded as 0; as shown in Figure 4. Finally, the comparison results are combined in order to form a 64-bit hash fingerprint. What needs to be explained here is that the hash fingerprints of the instant image and the template image need to be arranged in the same order, as shown in Figure 4, the fingerprints arranged in sequence from top to bottom are:

1110000010011000100000100010000010000000000000001000100000100000.

The average hash algorithm process includes:

1) Downsizing: Downsizing the instant image to a size of 8×8 with a total of 64 pixels.

2) Simplify color: convert the 8×8 instant image into a grayscale image, and convert it to 64 levels of grayscale.

3) Calculate the average: calculate the gray average of all 64 pixels; for example:

4) Compare the grayscale of the pixel: compare the grayscale of each pixel with the average value, if it is greater than or equal to the average value, it will be recorded as 1, and if it is less than the average value, it will be recorded as 0; as shown in Figure 5.

5) Generate a hash fingerprint: combine the comparison results of 4) in a certain order to form a 64-bit hash fingerprint. As shown in Figure 5, the fingerprints arranged in sequence from top to bottom are:

11111111111111111011111101101111000011110000111100001111100011.

6. Compare the hash fingerprint obtained from the real-time image with the hash fingerprint obtained from the template image. When the Hamming distance of the DCT-perceived hash fingerprint is greater than 10, and the Hamming distance of the average-perceived hash fingerprint is greater than 20, the video There are vehicles in the area, and the control lever remains vertical at this time.

7. The processing module receives the image information sent by the image acquisition module and repeats steps 4-6 until the Hamming distance of the DCT-perceived hash fingerprint is less than 5, and the Hamming distance of the average-perceived hash fingerprint is less than 10, then the brake lever is controlled to fall .

The present invention can reduce the application of ground induction coils in system composition, thus avoiding its defects in road induction; at the same time, the image is analyzed simultaneously through two hash algorithms, which can ensure the rationality of the algorithm and enhance the algorithm The anti-interference ability of the system can improve the reliability and stability of the algorithm and system.

Claims (4)

1. A vision-based driveway brake lever anti-collision system, comprising a barrier gate, a processing module, an image acquisition module, a memory module and a host computer; the image acquisition module is installed next to the barrier gate, and faces the target area; the The processing module analyzes the image information collected by the image acquisition module, and controls the state of the railing of the barrier; it is characterized in that: the memory module stores a template image, and the processing module receives the triggering of the barrier from the upper computer of the driveway. When the pole signal is used, the control memory module updates the image information collected by the image acquisition module into a template image; the processing module performs Gaussian filtering on the image information to obtain an instant image; Hash fingerprints, hash fingerprints are image description strings combined in the same order; when the Hamming distance between the hash fingerprints of the instant image and the template image is greater than the set value, the gate will keep the railing erect; When the clear distance is less than the set value, the barrier of the barrier will be lowered; The processing module generates the 64-bit hash fingerprint of the instant image and the 64-bit hash fingerprint of the template image by the DCT hash algorithm, and calculates the Hamming distance M of the two; the processing module generates the instant image by the mean value hash algorithm Calculate the Hamming distance N between the 64-bit hash fingerprint and the 64-bit hash fingerprint of the template image; if M>10 and N>20, keep the railings upright; if M<5 and N<10, The railing is down. 2. A kind of vision-based lane brake lever anti-collision system according to claim 1, characterized in that: the hash fingerprint generation step of the instant image or template image of the DCT hash algorithm is: 1) Reduce the image to a size of 32×32; 2) convert the reduced image into a grayscale image; 3) performing DCT transformation on the grayscale image, and obtaining a 32×32 DCT coefficient matrix; 4) Reserving the 8×8 matrix in the upper left corner of the DCT coefficient matrix; 5) calculation step 4) coefficient mean value in the DCT coefficient matrix; 6) Compare the coefficients in the DCT matrix in step 4) with the mean value one by one, if it is greater than or equal to the mean value, record it as 1; if it is less than the mean value, record it as 0; combine the comparison results in order to form a 64-bit hash fingerprint. 3. A kind of vision-based lane brake lever anti-collision system according to claim 1, characterized in that: the hash fingerprint generation step of the instant image or template image of the mean value hash algorithm is: 1) Reduce the image to a size of 8×8, a total of 64 pixels; 2) Convert the 8×8 image into a grayscale image, and convert it into a 64-level grayscale; 3) Calculation step 2) grayscale average value of all 64 pixels; 4) Compare the 64-level grayscale in step 2) with the grayscale average value one by one. If it is greater than or equal to the average value, record it as 1; if it is less than the average value, record it as 0; combine the comparison results in order to form 64-bit hash fingerprint. 4. A kind of vision-based lane brake bar collision avoidance system according to claim 1, characterized in that: the image acquisition module includes a CCD image sensor and a SAA7113 video decoding chip connected in sequence; the processing module adopts a DM642 chip ; The SAA7113 video decoding chip is connected to the VP1 interface of the DM642 chip, and the DM642 chip is connected to control the SAA7113 video decoding chip through an I2C bus.