CN112201056A

CN112201056A - Vehicle queuing length detection method based on angular point characteristic analysis

Info

Publication number: CN112201056A
Application number: CN201910608002.1A
Authority: CN
Inventors: 刘新平; 张影; 王风华
Original assignee: China University of Petroleum East China
Current assignee: China University of Petroleum East China
Priority date: 2019-07-08
Filing date: 2019-07-08
Publication date: 2021-01-08

Abstract

The invention relates to a vehicle queuing length detection method based on angular point characteristic analysis, and belongs to vehicle flow detection of an intelligent traffic management system. The method obtains an improved FAST algorithm by combining the traditional FAST corner detection method and the motion detection process, and can extract the corner feature map representing the existence of vehicles on the current traffic road and acquire the motion state of the corner position by using the improved FAST corner feature analysis technology. And obtaining the result of the improved FAST algorithm, forming vehicle queue by the static corner point characteristics in the single lane, carrying out PCA (principal component analysis) processing to obtain a low-dimensional vector, and finally carrying out morphological processing on the low-dimensional vector to detect the vehicle queue length in the single lane. The result shows that the detection precision of the method is 98% on average, and the method can be applied to actual scenes.

Description

Vehicle queuing length detection method based on angular point characteristic analysis

Technical Field

The invention discloses a vehicle queuing length detection method based on angular point characteristic analysis, and belongs to vehicle flow detection of an intelligent traffic management system.

Background

Road traffic parameters such as traffic flow, vehicle speed, lane occupancy and the like can be obtained through the detection result of the vehicle queuing length. The detection process of the vehicle queue length mainly comprises the steps of motion detection, vehicle existence detection and vehicle queue length measurement. At present, three methods are commonly used for a vehicle queue length detection algorithm, and the first method is a vehicle detection method combining background difference and vehicle characteristics. Yang Lianggyi et al use a three-frame time difference method of motion detection to detect moving vehicles, edge information as a vehicle presence flag, and a moving detection window to detect the vehicle queue length at the intersection. And finally, obtaining the vehicle queuing length by using a camera calibration technology. The algorithm mainly obtains the moving vehicles in the video through a time difference method, and the queuing length of the static vehicles in the case of traffic jam cannot be detected. The second is a vehicle detection method based on vehicle characteristics and perspective transformation, Wang Chuang et al obtains a bird's-eye view of a traffic road by establishing a world coordinate system and using inverse perspective transformation, identifies lane lines by using hough transformation based on highway structural constraint, and finally detects a vehicle contour by using a sobel operator. The method obtains a clearer vehicle queuing result through inverse perspective transformation, but the method is based on vehicle queuing formation and neglects judgment of the vehicle motion state. Thirdly, most of the conventional detection methods for the vehicle queuing length are based on a sliding window method, and after the vehicle is extracted by Liu Zhe based on morphological edges, a vehicle queuing length detection method for a telescopic window is provided. The above algorithms can effectively detect the vehicle to a certain extent, but all have certain limitations.

Disclosure of Invention

On the basis of the existing vehicle queue length detection method, the invention mainly performs the following work in the vehicle queue length detection process. Firstly, the process of improving the FAST corner detection is to combine FAST algorithm detection with an interframe difference method, use corner information as a mark of vehicle existence in a detection window, and compare pixel change of corner positions between adjacent frames with a threshold value to judge the motion state of the vehicle and acquire a vehicle queue in a video. Secondly, when the current vehicle is in a static state, a static vehicle corner feature map is obtained, corner information in the queue is compressed into a low-dimensional vector by combining a PCA technology, and useful corner information in original video monitoring is retained while isolated noise points are removed. Thirdly, the obtained low-dimensional vector is subjected to morphological processing of a variable template, and finally converted into the actual queuing length. In order to achieve the purpose, the technical scheme of the invention is as follows:

step one, preprocessing the real-time road condition video collected by a traffic intersection camera. The method comprises the steps of graying a video image, removing noise and extracting an interested area;

step two, judging the position and the motion state of an angular point in a current detection window by using an improved FAST algorithm so as to judge the formation of vehicle queuing;

step three, acquiring a vehicle queue when a large number of static angular point features are detected in the current detection window;

step four, the corner features obtained in the queue are compressed into low-dimensional vectors by applying PCA (principal component analysis) processing to represent the length of the vehicle queue;

step five, performing morphological processing on the PCA result to determine the queue length;

and step six, converting the result obtained in the step five into the pixel length.

Has the advantages that:

the method for detecting the vehicle length combines PCA processing with FAST algorithm, and is characterized in that a low-dimensional data set generated by the PCA processing can retain the variable of original data as much as possible. And (3) generating a new image by using an improved FAST algorithm and the corner feature key point set according to the mapping principle, so that the image is conveniently used for PCA processing to obtain a low-dimensional vector to detect the queuing length of the vehicle.

In the existing vehicle queue length detection method, the indispensable steps comprise motion detection and vehicle existence detection before vehicle queue detection, and the second point of improvement of the FAST algorithm in the method is that the motion detection method is integrated into the FAST algorithm, so that the operation burden of the system is reduced.

Drawings

FIG. 1 is a flow chart of a vehicle queue length detection method according to an embodiment of the invention;

FIG. 2 is a schematic diagram of region of interest extraction according to an embodiment of the present invention;

FIG. 3 is the result of the modified FAST algorithm within the detection window in the example of the present invention;

FIG. 4 is a static corner feature map in an example of the present invention;

FIG. 5 is a compression result of the PCA algorithm in an example of the present invention;

FIG. 6 shows the results of morphological processing in an example of the invention;

FIG. 7 is a pixel length of vehicle queue in an example of the invention.

Detailed Description

The invention discloses a vehicle queuing length detection method based on angular point characteristic analysis. The method is based on FAST corner detection to extract the vehicle corner features in a single lane in a surveillance video, and uses PCA to extract principal components and compress the principal components into low-dimensional vectors. And finally, obtaining the vehicle queuing length by using the morphological processing of the variable template.

In the following detailed example, referring to fig. 1, the vehicle queue length detection based on the corner feature analysis is further explained:

in this embodiment, the vehicle queuing length at a certain intersection is detected, and the detection steps are as follows:

example (b):

step one, preprocessing the real-time road condition video collected by a traffic intersection camera. Including graying the video image, noise removal and region of interest extraction. The extraction result of the region of interest in different scenes is shown in fig. 2;

step two, after preprocessing the original video, setting a virtual detection area in front of the stop line, judging the position and the motion state of the corner point in the current detection window by using an improved FAST algorithm, and distinguishing the static corner point from the dynamic corner point by using different colors, as shown in fig. 3. Assuming that p is the position where the corner is detected, the method for detecting the motion state is as follows:

it (p) is the value of the pixel p at time t, s (p) is the set of neighborhoods of p, and comparing the obtained d (p) with the threshold t to obtain e (p), i.e. the motion state of the vehicle. When E (p) is 1, the static corner points are, and when E (p) is 0, the dynamic corner points are;

and step three, acquiring a vehicle queue when a large number of static corner point features are detected in the current detection window, as shown in fig. 3 (b). When the vehicles in the virtual detection area are in a static state, the default traffic is in a congestion state and the vehicle queue length is measured. The method generates a new image from the static corner feature key point set according to the mapping principle, and is convenient to use the image to carry out PCA processing to obtain a low-dimensional vector, wherein the image is as shown in FIG. 4;

and step four, PCA processing is applied to the corner features obtained in the queue, and the corner features are compressed into low-dimensional vectors to represent the length of the vehicle queue. The low-dimensional vector map shown in fig. 5 is generated by performing PCA on the corner point map generated as shown in fig. 4;

and step five, performing morphological processing on the PCA result to determine the queue length. The above obtained discontinuous results are optimized using morphological treatments. Due to the large and small features during imaging, as shown in fig. 6, the image is processed using variable template morphology:

where B (i, j) represents the range of the matrix formed by aligning the center of B with the (i, j) position of Ar, which is an image expanded using the variable template B (2r +1 ). Ar (p, q) is the point falling on the B (i, j) # Ar portion, resulting in its maximum value;

and step six, converting the obtained morphological processing result into a pixel length, wherein the vehicle queue length in the invention is formed by white areas, namely, a first line white area encountered from (x, y) ═ 0, 0+ +, is the tail of the vehicle queue, and a first line white area encountered from (x, y) ═ 0, img. The pixel length obtained in this example is as shown in fig. 7.

Claims

1. A vehicle queuing length detection method based on angular point feature analysis is characterized by comprising the following steps:

and step two, after the original video is preprocessed, a virtual detection area is arranged in front of a stop line, the position and the motion state of the corner point in the current detection window are judged by using an improved FAST algorithm, and static corner points and dynamic corner points are distinguished by using different colors. Assuming that p is the position where the corner is detected, the method for detecting the motion state is as follows:

it (p) is the value of the pixel p at time t, s (p) is the set of neighborhoods of p, and comparing the obtained d (p) with the threshold t to obtain e (p), i.e. the motion state of the vehicle. When E (p) is 1, the static corner is, and when E (p) is 0, the dynamic corner is;

and step three, when the vehicles in the virtual detection area are in a static state, defaulting that the traffic is in a congestion state, acquiring a vehicle queue, and measuring the vehicle queue length. The method generates a new image from the static angular point feature key point set according to the mapping principle, and is convenient for using the image to carry out PCA processing to obtain a low-dimensional vector;

step four, the corner feature maps obtained in the queue are compressed into low-dimensional vectors by applying PCA (principal component analysis) processing to represent the length of the vehicle queue;

and step five, performing morphological processing on the PCA result to determine the queue length. The above obtained discontinuous results are optimized using morphological treatments. Due to the characteristics of the near size and the far size in the imaging process, the image is subjected to morphological processing by using a variable template:

step six, the vehicle queue length in the invention is composed of white areas, namely, the white area of the first line encountered from (x, y) ═ 0, 0+ +, is the queue tail of the vehicle queue, and the white area of the first line encountered from (x, y) ═ 0, img.

2. The method for detecting vehicle queue length based on corner feature analysis according to claim 1, characterized in that FAST algorithm detection is combined with frame-to-frame difference method to determine the motion state of the corner while detecting it.

3. The vehicle queue length detection method based on corner feature analysis as claimed in claim 1, characterized in that the improved FAST algorithm obtains the static vehicle corner feature map, and compresses the corner information in the queue into low-dimensional vectors by combining PCA technology.

4. The method of claim 1, wherein the low-dimensional vector obtained by the PCA is subjected to a variable template morphology process.