Research On The Detection And Tracking Algorithm Of Underground Personnel Target

There are some dangerous areas in the coal mine, these areas are not allowed to staff in without the protection measures to enter in. But due to some environmental and artificial factors, make the warning effect of these general identifications not strong enough. In order to achieve safety production, it is necessary to take some effective detection means. Due to the artificial lighting the way, the lack of color information, and the color of the target and the background color is similar, so it will face great difficulties when use intelligent monitoring technology in underground coal mine. In this paper what can be used in underground environment of target detection and tracking methods are studied, specific work is as follows:(1) Mixture Gaussian Modeling method is one of the widely used detection methods, but the traditional gaussian background model process to establish a fixed number of Gaussian models and the same learning rate separately for each pixel, these defects were improved in this paper. To block the read in video frames first, and then adaptively for each pixel block assign a different number vector and the gaussian distributions, at the same time of airspace improvement, in the time domain and modeling process of different stages for different vector distribution, make the modeling more rapid and accurate. Simulation results show that the improved algorithm in the time domain and the airspace two aspects improve the target detection accuracy at the same time satisfy the real-time requirement.(2) Due to the own defects of mean shift algorithm for target tracking and environment constrains, this paper respectively from the modeling characteristics of select, update of the bandwidth of kernel function and the judgment of the sports area carried on the optimization. When select the modeling feature, combine the LBP texture with the color feature which used in the original algorithm embedded to the mean shift tracking framework, and describe the goal. Use frame difference method to test the change region between the adjacent frames, to treat the movement of the center of area as the initial position of iteration, shortening the distance between the initial position and the true position of the target, thus reducing the number of iterations. In terms of the size of the kernel function of bandwidth, by combining positioning location information of the boundary of the target area to realize the adaptive updating, for target tracking in the integrity of the information at the same time eliminate the interference of background pixels as far as possible, makes the tracking more rapid and accurate.