WO2021114799A1

WO2021114799A1 - Computer vision-based matrix vehicle light identification method

Info

Publication number: WO2021114799A1
Application number: PCT/CN2020/116242
Authority: WO
Inventors: 王建春
Original assignee: 华南理工大学广州学院
Priority date: 2019-12-14
Filing date: 2020-09-18
Publication date: 2021-06-17
Also published as: CN111027494A; CN111027494B

Abstract

A computer vision-based matrix vehicle light identification method. The method comprises the steps of: obtaining pavement vehicle image video information by means of a camera; performing image binarization processing; performing image edge detection; obtaining vehicle light distance information; sending processed distance information to a lower computer; and adjusting, by the lower computer, brightness of a vehicle light according to the distance information. According to the method, a pavement vehicle image is automatically obtained, the image is subjected to analysis processing, and brightness of a vehicle light is adjusted according to the processing result, so that the vehicle light is automatically adjusted according to a pavement vehicle condition, and traffic accidents are prevented.

Description

A matrix car lamp recognition method based on computer vision

Technical field

The invention relates to the technical field of automobile lights, in particular to a matrix automobile lamp recognition method based on computer vision.

Background technique

With the improvement of people's living standards, cars are an indispensable means of transportation for people to travel. As people's demand for cars has increased, their requirements for driving comfort and functions have also been increasing. For car light control, a civilized and good driving lighting program is needed. At present, most of the automotive matrix lamp control technology relies on photosensitive sensors, ultrasonic sensors, etc., which can only detect information such as environmental brightness in the surrounding environment of the car, and the ability to judge such information is limited. The reason is that the night road image contains A lot of interference information, such as street lights, reflective road signs, and reflective road surface water, will interfere with the identification of the lights during the night meeting, which cannot meet the further requirements of vehicle identification, causing great security risks.

In order to solve the above-mentioned technical problems, a method for controlling the high beams of a night meeting car has appeared. A patent document with a Chinese patent application number of 201810810013.3 and an announcement date of 2018.12.28 discloses a method for controlling the high beams of a night meeting car. The system, computer-readable storage medium, and the method include the steps of: acquiring a road vehicle image; acquiring the brightness value of the HSL image pixel according to the road vehicle image; judging whether the brightness value of the HSL image pixel is within a preset vehicle light brightness threshold range If the judgment result is yes, then match the road vehicle image with the preset image mask; if the match is successful, obtain the location information of the road vehicle based on the road vehicle image; control based on the location information of the road vehicle The high beams of the corresponding area are turned off.

However, according to the technical solution disclosed in this document, the road vehicle image is matched with the preset image mask, specifically by matching the brightness value, and then the vehicle lights are adjusted after the matching is successful. However, because different vehicles have different brightness and There will also be deviations in the brightness collected from different locations, and the brightness of the car lights detected on some roads will also be affected by the brightness of one side of the road, which makes it impossible to truly judge whether the vehicle is getting closer and closer to the vehicle. This leads to inaccurate control.

Summary of the invention

The invention provides a matrix vehicle lamp recognition method based on computer vision. The method has high accuracy in vehicle distance recognition, and adjusts the brightness of the vehicle lamp according to the distance from the road vehicle when meeting vehicles at night to prevent the occurrence of safety accidents.

In order to achieve the above-mentioned purpose, a matrix vehicle light recognition method based on computer vision includes the following steps:

1) Obtain the image and video information of road vehicles through the camera;

2) Carry out image binarization processing on the vehicle image in the vehicle image and video information;

3) Preprocess the image: remove image noise through Gaussian filtering to obtain a denoised image;

4) Perform image edge detection: detect whether there are lights of the target car on the image, if it is detected that the image after denoising has two areas of similar size and the distance between the two areas is below the threshold range, then the image after denoising is determined If there are car lights in the image after denoising, calculate the area of the connected area according to the detection result to obtain the largest area of the connected area, draw the outline and obtain the outline of the largest area; use the Hough circle to detect the largest area. Calculate the largest closed circle on the area contour; obtain the ROI rectangular area through the calculated center and radius parameters of the largest closed circle, and use the obtained ROI area as the vehicle light area to be tracked in the next step;

5) Obtain the distance information from the target car: back-project the car light area identified in 4) to obtain the back-projection view; perform the camshift algorithm iteration according to the obtained back-projection map and the outline of the car light, according to the previous Frame image and the center coordinates of the ROI rectangular area of the next frame image iteratively output by the camshift algorithm; through the formula AB=((X+P2)*F)/K, the distance to the target car is obtained, where AB is the distance to the target car X is the center coordinate of the image output of the previous frame, P2 is the center coordinate of the image output of the current frame, F is the hyperparameter, and K is the adjustment parameter;

6) Send the processed distance information to the lower computer;

7) The lower computer adjusts the brightness of the lights according to the distance information from the target vehicle. The lower computer adjusts the voltage of the LED lights through the output voltage of the control circuit, thereby adjusting the brightness of the lights.

With the above settings, the road vehicle image and video information are automatically obtained through the camera, and then the vehicle image is binarized, and the image is denoised, and then the edge detection is determined to determine whether there are lights, and then the outline of the lights is determined, and at the same time pass Edge detection determines the largest closed circle and then determines the ROI area, that is, the key area, and then determines the area before and after the movement of the lamp, and then determines the distance between the two cars according to the difference between the area parameters. Since the detected car and the target car are moving, according to the detected The area difference between the previous frame image and the current frame image of the vehicle image can be judged as the distance between the detection vehicle and the target vehicle, and then the brightness of the detection vehicle is adjusted according to the distance to ensure the safety of driving at night. The movement distance determined in this way will not be affected by the brightness of the external environment, and the accuracy is high. After edge detection, the maximum closed circle is determined and then the ROI is determined, which can save data processing and also determine the key information area through ROI to improve accuracy .

Further, the image edge detection in 4) is Laplacian image detection. The above setting method is simple and reliable.

Further, the lower computer adjusts the brightness of the vehicle lights according to the processed distance information. The closer the distance, the lower the brightness of the vehicle lights, so as to prevent the high brightness of the vehicle lights from causing dazzling situations to road vehicle drivers and causing traffic accidents. With the above settings, the closer the distance is, the higher the brightness is, the more dangerous it is. Therefore, the brightness information is adjusted according to the distance information to ensure driving safety when meeting cars at night.

Further, said 7) the lower computer adjusts the brightness of the car lights according to the distance information, which specifically includes the distance information and the preset relationship table between the distance information and the brightness to determine the brightness information, and the brightness information is determined according to the distance information and the preset relationship between the distance information and the brightness. The meter can adjust the brightness more accurately and improve the accuracy of the brightness adjustment of the car lights.

Further, in step 2), performing image binarization processing on the vehicle image specifically includes: converting the vehicle image to RGB color space, and then performing binarization processing on the RGB color space, by converting the vehicle image to RGB color space, and then Carrying out binarization processing, processing is convenient, and processing speed is fast.

Description of the drawings

FIG. 1 is a schematic flowchart of a method for matrix vehicle light recognition based on computer vision of the present invention.

Fig. 2 is a picture after denoising processing of an image in a matrix car light recognition method based on computer vision of the present invention.

FIG. 3 is a picture after edge detection is performed on an image in a matrix vehicle light recognition method based on computer vision of the present invention.

Detailed ways

The present invention will be further described in detail below in conjunction with the drawings and specific embodiments.

The present invention discloses a matrix vehicle lamp recognition method based on computer vision, as shown in Fig. 1, including the following steps:

1) Obtain the image and video information of road vehicles through the camera;

4) Perform image edge detection: As shown in Figure 3, detect whether there are lights of the target car on the image. If it is detected that the denoising image has two areas of similar size and the distance between the two areas is below the threshold range, then Determine that there are car lights in the image after denoising; if it is detected that there are car lights on the image after denoising, calculate the area of the connected region according to the detection result to obtain the largest connected region area, draw the contour and obtain the contour of the largest area; use Huo Husband circle detection, find the largest closed circle on the contour of the largest area; obtain the ROI rectangular area through the obtained center and radius parameters of the largest closed circle, and use the obtained ROI area as the vehicle light area to be tracked in the next step; this embodiment In the Hough circle detection, the Hough circle transformation in OpenMV is used.

5) Obtain the distance information from the target vehicle: Obtain the distance information from the target vehicle: Back-project the lamp area identified in 4) to obtain the back-projection view; according to the obtained back-projection diagram and the outline of the lamp Carry out camshift algorithm iteration, according to the previous frame image ROI area area S1 and the current frame image ROI rectangular area area S2 iteratively output by the camshift algorithm; through the formula AB=(S2-S1)/K, the distance to the target car is obtained , Where AB is the distance between the target vehicle and K is the adjustment parameter; the area of the ROI rectangular area can be calculated by determining the center and radius parameters of the largest closed circle due to the Hough circle detection, and the ROI rectangular area is the largest closed circle The largest inscribed circle of a circle, since the radius parameter of the inscribed circle is determined, the area can be obtained by 4 times the square of the radius parameter;

Determine the distance between the two cars after adjusting the difference between the ROI area area S1 of the previous frame image and the ROI rectangular area area S2 of the current frame image;

6) Send the processed distance information to the lower computer;

In this embodiment, the image edge detection in 4) is Laplacian image detection.

The lower computer adjusts the brightness of the car lights according to the processed distance information from the target car. The closer the distance to the target car, the smaller the brightness of the car lights. Specifically, according to the distance information from the target car and the preset distance target car The relationship table between the distance information and the brightness determines the brightness information, so as to adjust the brightness of the lights according to the distance information of the target vehicle, and prevent the brightness of the lights from being too high to cause dazzling conditions for road vehicle drivers and cause traffic accidents.

In this embodiment, performing image binarization processing on the vehicle image in step 2) specifically includes: converting the vehicle image to an RGB color space, and then performing binarization processing on the RGB color space;

Automatically obtain the video information of the road vehicle image through the camera, and then binarize the vehicle image, and perform the denoising process on the image, and then determine the edge detection to determine whether there are lights, then determine the outline of the lights, and determine through edge detection The largest closed circle then determines the ROI area, that is, the key area, and then determines the area before and after the movement of the lights, and then determines the distance between the two cars according to the difference between the area parameters. Since the detected car and the target car are in motion, according to the detected vehicle image The area difference between the previous frame image and the current frame image can be determined as the distance between the detection vehicle and the target vehicle, and then the brightness of the detection vehicle is adjusted according to the distance to ensure the safety of driving at night, and it is determined according to this method The moving distance will not be affected by the brightness of the external environment, and the accuracy is high.

Claims

A method for matrix vehicle lamp recognition based on computer vision, which is characterized in that it comprises the following steps:

1) Obtain the image and video information of road vehicles through the camera;

2) Carry out image binarization processing on the vehicle image in the vehicle image and video information;

3) Preprocess the image: remove image noise through Gaussian filtering to obtain a denoised image;

4) Perform image edge detection: detect whether there are lights of the target car on the image, if it is detected that the image after denoising has two areas of similar size and the distance between the two areas is below the threshold range, then the image after denoising is determined If there are car lights in the image after denoising, calculate the area of the connected area according to the detection result to obtain the largest area of the connected area, draw the outline and obtain the outline of the largest area; use the Hough circle to detect the largest area. Calculate the largest closed circle on the area contour; obtain the ROI rectangular area through the calculated center and radius parameters of the largest closed circle, and use the obtained ROI area as the vehicle light area to be tracked in the next step;

5) Obtain the distance information from the target car: back-project the car light area identified in 4) to obtain the back-projection view; perform the camshift algorithm iteration according to the obtained back-projection map and the outline of the car light, according to the previous Frame image ROI area area S1 and the area S2 of the current frame image ROI rectangular area iteratively output by the camshift algorithm; through the formula AB=(S2-S1)/K, the distance to the target car is obtained, where AB is the distance to the target car The distance between the two, K is the adjustment parameter;

6) Send the processed distance information to the lower computer;

7) The lower computer adjusts the brightness of the lights according to the distance information from the target vehicle. The lower computer adjusts the voltage of the LED lights through the output voltage of the control circuit, thereby adjusting the brightness of the lights.
The method for matrix vehicle light recognition based on computer vision according to claim 1, wherein the image edge detection in 4) is Laplacian image detection.
The method for matrix vehicle light recognition based on computer vision according to claim 1, wherein the lower computer adjusts the brightness of the vehicle lights according to the processed distance information from the target vehicle, the closer the distance to the target vehicle is, The lower the brightness of the car lights.
The method for matrix vehicle light recognition based on computer vision according to claim 3, characterized in that: said 7) the lower computer adjusts the brightness of vehicle lights according to distance information, specifically including distance information based on the distance to the target vehicle and presets The relationship table between the distance information of the target vehicle and the brightness determines the brightness information.
The method for matrix vehicle light recognition based on computer vision according to claim 1, characterized in that: in step 2), the image binarization processing on the vehicle image specifically includes: converting the vehicle image to RGB color space, and then The RGB color space is binarized.