Research And Implementation On Target Tracking Algorithm Of The Underground Personnel With Low Illumination

Currently coal industry has basically equipped with video surveillance systems,but most of the underground video surveillance system only has manual work to monitor the subsurface environment. In the course of a long period of surveillance,monitoring personnel can easily fatigue,which causing accidents. If we can use intellectual detection and tracking technology on the basis of the existing video surveillance system to get the real-time location information of underground personnel and send some security warnings or instructions,we will be able to improve the safety of coal mine operations.However,different from ordinary scenes, scenes of coal mine has low overall illumination, uneven illumination at each position, and there is no color information like traditional video, which resulting in great similarity of target and background,these problems not only make the target detection of underground scene target detection very difficult, but also lead to the traditional tracking methods unsatisfactory. In order to improve the situation of underground video target tracking, this paper studied and analyzed the features of the underground scene, and improved the process and algorithm of traditional video tracking, and its main contents are as follows:1.This paper use mixture Gaussian background modeling to dynamically update the background of the scene, so that to reduce the impact on the target detection of gradually background changing. And in order to solve the problems caused by light spot and shadow,this paper digest the information of the mining scene, and proposed a method which combined the correlation coefficient method to eliminate light spots and shadows, and made some experiment and comparative experiments on video sequence which contents problems of light spots and shadows to verify the effectiveness of the proposed algorithm.2.Considering the characteristics of the underground scene, the paper chose HOG feature as the main characteristics of the target tracking, and use particle filter tracking algorithm to do the tracking.The paper improved the traditional particle filter algorithm by using discriminant method to measure the particles which leads to a better distribution and better tracking result. In the actual tracking process the paper use coredensity estimation to calculate the probability if the foreground image belongs to the target or the background,so that to get the real human body target.The paper improve the tracking accuracy by establishing real-time updates of sample libraries and adding removal light shadow removal. The paper use the Kalman-Camshift and SVM tracking algorithm to do the contrast test, the results of experiment show that the proposed tracking algorithm in the underground scene has better adaptability and is effective.