Research And Implementation Of Object Detection System Based On Optimized ViBE And HOG

With the rapid development of computer technology and the widely use of digital image technology, computer vision has been very popular. Recently object detection is very popular. Object detection is a kind of image segmentation based on statistical characteristics and object geometry that merge object segmentation and recognition described by accuracy and real-time performance. The most important application of motion object detection is extracting interest region from the video image sequence. An important standard for evaluation of target recognition system is that it can obtain accurate results through the analysis of the camera image sequence processing without artificial auxiliary.It’s main goal of this study is to optimize motion object detection in video from a fixed camera. Main work of this article is embodied in three aspects:(1)Optimization algorithm for HOG template matching method. At present, HOG feature is described to be a very effective way to detect a object that includes two steps: feature extraction and template matching. The first step for making template is extracting HOG feature from the data set, and then training classifier to be template. For two kinds of objects with shape similarity, the trained template outline may be more similar. The score may be not too high if the object appears to be detected or deformation caused by the appearance interference. Then the score of real template is near to similar templates. For solving this kind of situation, this article introduce deformable part model for further calculation and confirmation.(2) In terms of video positioning, ViBE algorithm is introduced in this paper. The Vi BE algorithm have some problems such as ghost. We put forward the ghost elimination method and parameter adaptive algorithm in this context for improving ViBE algorithm accuracy.(3)We propose object detection system based on optimized ViBE and HOG with the respective characteristics of them in fixed cameras video.Through a lot of experimental analysis we can find that this algorithm improve the accuracy with the same velocity.