A fine vehicle type identification and flow statistics method based on deep learning and trajectory tracking

Abstract

The invention provides a fine vehicle type identification and flow statistics method based on deep learning and trajectory tracking. The method comprises the steps of obtaining vehicle images and establishing a sample set; Marking the position and type of the vehicle, the position of the wheel and the type information of the wheel in each picture; According to the vehicle images in the sample setand the corresponding annotation information, training an established deep learning model; Utilizing the deep learning model to predict the position and type of the vehicle, the position of the wheeland the type information of the wheel in the image; Estimating the actual length and height of the vehicle and the number of side wheels in the video according to the pinhole imaging model, and performing fine classification on the vehicle type; And obtaining a motion track of each vehicle in the video by utilizing a tracking and matching algorithm, and counting the flow of different types of vehicles according to the motion tracks.

Description

technical field [0001] The invention relates to the field of automobile monitoring, and mainly relates to a low-cost, fine vehicle identification and flow statistics method based on deep learning and trajectory tracking. Background technique [0002] With the vigorous development of expressway passenger and freight transportation in my country, highway safety accidents are also constantly occurring, and expressway passenger car and truck accidents often cause road congestion, huge casualties and other problems, so the safety of expressway passenger and freight transportation has become the focus of current attention. It is of great significance to the refined model identification, monitoring and traffic statistics of passenger cars and trucks. [0003] The current car model recognition algorithm based on high-speed video surveillance can identify few car models, and generally only realize the recognition of family cars, buses, and trucks, but lacks refined car model recogniti...

AI Technical Summary

Problems solved by technology

[0004] The current algorithm has a single function, and generally only can identify vehicle types or total traffic statistics, and cannot integrate multiple functions to track and monitor vehicles in an all-round way;

[0005] In addition, the current method focuses on fine-grained car model identification for family cars, which is of little significance in highway monitoring scenarios

Images