Vehicle queue length fast detection algorithm based on local feature analysis

Abstract

The invention relates to a vehicle queue length fast detection algorithm based on local feature analysis, which belongs to vehicle flow detection in intelligent traffic. In a video monitoring area, local features of an image are optimized from the whole image, three columns of pixel values containing a lane image are only selected when the traffic information of the vehicle queue length is detected, and a one-dimensional feature array is formed on the basis of weighted reconstruction for analysis. The imaging principle of a camera is simplified into a pin hole model, thereby realizing conversion between a pixel distance and an actual distance; Gaussian filtering and wavelet transform are used to be combined to reduce noise interference, the wavelet base section is db4, and the number of decomposition levels is 3; a local background subtraction method is used for extracting foreground; and finally, a sliding window with a variable length is used for detecting a queue tail. The algorithm of the invention is used for testing a photographed traffic monitoring video, the result shows that the algorithm accuracy is high, the operation rate is fast, errors in the field of vision is smaller than 5%, time consumed for single frame processing is only 10ms, and actual application requirements can be met.

Description

technical field [0001] The invention relates to a fast detection algorithm of vehicle queuing length based on local feature analysis, which belongs to the detection of vehicle flow in an intelligent traffic management system. Background technique [0002] At present, a variety of sensors have been used to detect the presence, quantity, queuing situation, and average speed of vehicles, among which vehicle detection based on machine vision is the most widely used. Vehicle queue length detection based on machine vision usually consists of the following steps: vehicle presence detection, tail vehicle detection, projection transformation, etc. The detection algorithm based on background difference is the most commonly used, which relies to a certain extent on background modeling with complex calculations and large system resource consumption. The purity of the background also directly affects the accuracy of detection results. The algorithm mentioned by Albiol A et al. uses the ...

AI Technical Summary

Problems solved by technology

This algorithm improves the detection range at the cost of sacrificing distance accuracy, and cannot accurately locate traffic intersections, and draws a conclusion for an intersection

Song Xiaona combined the AdaBoost algorithm with a simple convolutional neural network to achieve fast classification and real-time recognition, but the algorithm has high complexity and a large amount of code, which is not suitable for use in embedded systems with limited resources

The above algorithms can effectively detect vehicles to a certain extent, but they also have certain limitations.

Images