Algorithm Optimization And Implementation On Surveillance Video Labeling On CUDA And OpenCV Hybrid Platforms

The development of vision technology makes surveillance video being widely used in various fields including the campus. With the increasing huge amount of information of Campus surveillance video, it is particularly important to extract meaningful information and label them. Rely on the powerful calculating ability of GPU, the CUDA platform researched and spread by NVDIA company follows the trend, become the leading technology support platform in the field of video data processing.This paper is based on the above background, takes full advantage of the CUDA platform, takes marking the moving target in video as a target.While completing the design,there is a brief analysis of parallel algorithm improving efficiency in this paper. This paper integrate the campus video monitor and parallel computing, in order of moving target track and dynamic critical information labeling. The paper designs a simple campus surveillance video labeling system based on CUDA and OpenCV hybrid platforms. The video information labeling system uses the method based on the CUDA parallel computing,uses background difference method to be the core algorithm of moving target detection, target has processed successively by the key steps such as background subtraction, two value processing and expansive-corrosion treatment.it uses the method of parallel degree reduction to improve the efficiency of algorithm, at the same time,it uses Qt framework to realize the system client, finally realizes the contour extraction and annotation of the image.The main tasks include:testing of moving target detection based on parallel computing; contour extraction and labeling based on OpenCV; Related data conversion under CUDA and OpenCV hybrid platform; data transmission based on Socket multithreaded network; Simple multithreading display and functional interface realization based on Qt. Through experimental analysis, results demonstrate that this moving target track and labeling systems can correctly detect the moving target and make out markers, it proves the advantage of parallel algorithm in the efficiency of video processing through the result of comparing lots of experimental data between the parallel algorithm and the thread algorithm.,it makes a try for video processing research in parallel computing method.