Parallel high definition video vehicle detection method based on GPU

Abstract

The invention relates to a parallel high definition video vehicle detection method based on GPU. The method comprises the following steps: a, an original video stream output by a high definition monitoring camera is transmitted into a display memory of a display card, and GPU in the display card carries out parallel decoding on the original video stream to get a frame gray image; b, GPU divides each frame gray image into N*N image blocks, and carries out parallel extraction on SIFT characteristics of each image block; c, GPU carries out parallel sparse coding on each image block; d, a constraint matrix U corresponding to each image block is acquired by parallel calculation; e, a characteristic vector zk of each image block is parallelly calculated; f, GPU carries out parallel linear classification and identification on the above characteristic vector zk obtained by calculation to obtaina corresponding judgment result fk; and g, GPU returns the judgment result fk into CPU, and CPU controls the working state of the high definition monitoring camera according to the judgment result fk. The method provided by the invention can meet the requirement of real-time monitoring of the high definition video, and has the advantages of wide suitability, safety and reliability.

Description

technical field [0001] The invention relates to a vehicle detection method, in particular to a GPU (Graphic Processing Unit)-based parallel high-definition video vehicle detection method, belonging to the technical field of high-definition video vehicle detection. Background technique [0002] With the development of technology, in modern intelligent transportation systems, the vehicle detection and recognition method based on surveillance video has become the mainstream means and development trend of vehicle detection in intelligent transportation systems. How to complete high-accuracy vehicle detection in real time is a hot and difficult point in current research. [0003] At present, in the field of pattern recognition, the relatively mature technology is SVM to complete classification detection and various improved algorithms. Among them, we found that the algorithm based on sparse coding and linear SVM classifier has achieved very good results. [0004] The algorithm ...

AI Technical Summary

Problems solved by technology

[0005] The calculation speed of the algorithm mentioned above has been significantly improved compared with the nonlinear SVM (support vector machine). For high-definition video surveillance, its calculation speed is still difficult to meet the real-time requirements, and it can only judge whether the target object appears in the video. , the location information of the target cannot be given

Images