Research And Implementation Of Road Traffic Flow Detection Algorithm Based On Image

With the development of intelligent transportation systems and the need of front-line intelligence for road monitoring by the transportation department,the research and implementation of image-based road traffic detection algorithms has become an urgent need.Road traffic detection provides basic decision data for intelligent transportation systems.It helps the traffic management department to optimize the traffic scheduling,so the road traffic flow detection has very important theoretical significance and potential application value.On the basis of moving target detection algorithm,this paper analyzes and improves several problems existing in Gaussian Mixture Model and Multi-task Cascaded Convolutional Networks model in vehicle detection.A strategy of fusion Gaussian Mixture Model and Multi-task Cascaded Convolutional Networks are proposed.Finally,a traffic flow detection algorithm capable of high-precision detection of road traffic in a variety of complex environments,and successfully moved to the Hi3516 D platform.The main research contents of this paper are:(1)Aiming at the unreasonable allocation of Gaussian distribution components in the background modeling of Gaussian Mixture Model,only the isolated point modeling ignores the relevance of each point in the neighborhood of pixel points,and cannot accurately detect slow motion or short-term stationary moving targets.Adapted to Gaussian distributed component strategy,time domain spatial hybrid modeling strategy,two-way dynamic update learning rate modeling strategy,three improved schemes;accelerated background modeling speed,and realized slow moving targets and temporary stationary targets in more complex environments Accurate detection.At the same time,this paper proposes a bump matching target detection segmentation algorithm to solve the problem of vehicle mutual adhesion,and proposes a traffic flow statistics method based on Kalman trajectory tracking to realize the detection of road traffic flow;and designed multiple sets of experiments to prove the subject.Improve the effectiveness of the Gaussian Mixture Model vehicle traffic detection algorithm.(2)The problem of low accuracy of traffic flow detection in mixed Gaussian Mixture Model under road congestion conditions;a vehicle detection strategy based on congestion discriminant method for hybrid Gaussian model and Multi-task Cascaded Convolutional Networks model is proposed;The detection accuracy of the Multi-task Cascaded Convolutional Networks model is not high.The internal cascade structure and the dual-stream convolutional neural network structure are proposed.Finally,the accurate detection of road traffic flow under congestion conditions is realized.At the same time,this project designed a number of experiments to prove the vehicle traffic flow detection based on fuzzy comprehensive model congestion discrimination method,improved Multi-task Cascaded Convolutional Networks vehicle detection method,fusion Gaussian Mixture Model and Multi-task Cascaded Convolutional Networks model.(3)Based on the analysis of Hi3516 D platform characteristics,this project successfully implements the above traffic flow detection algorithm transplanted into Hi3516 D,and uses Hi3516 D Intelligent Video Engine to realize the speed optimization of traffic flow detection algorithm.The main process includes the construction of Hi3516 D development environment,the traffic detection algorithm relies on the transplantation of the library,the algorithm is divided and transplanted according to the process of processing the video sequence by Hi3516 D,and the algorithm is replaced and optimized by the hardware acceleration operator of the platform.A number of experiments were designed to prove the feasibility and effectiveness of the vehicle traffic detection algorithm transplanted to the Hi3516 D platform.