Research On Multi-object Tracing Algorithms Based On Moving Regions’ Features

This paper researched on multi-object tracking algorithms based on region features. Considering the computation of extracting and matching the whole image’s features to achieve the goal of multi-object tracking is huge, and the background features are usually unnecessary, especially for the complicated background which contains many features that do not contribute to object matching. To reduce the unnecessary computation of feature extracting and matching, this paper proposed a method of matching the regions’ features to track the objects. The experimental results show that the proposed method can reduce one of tenth computation.This paper researched on ViBe motion detection algorithm and implemented the ViBe algorithm to extract the moving regions of the image frame. With the ViBe algorithm implemented an optimized in this paper, the moving regions of the image frame can be extracted correctly. The moving regions are prepared for feature extracting and matching.After extracting the moving regions, SIFT algorithm is applied to extract the features in order to match with object feature database. The procedure of the proposed method is as follow: 1, build up the object feature database; 2, for each image frame, extract the SIFT features of the moving regions; 3, match the features of the moving regions and the object feature database to estimate whether the objects to be tracked appear in the image. This paper also proposed 2 parameters, the feature group number, , and feature matching threshold, ???, for feature matching to estimate the appearance of objects in the image. The experimental results show that the proposed method can achieve relatively good accuracy of multi-object tracking.The proposed method can be applied in the field of smart surveillance to auto-recognize the objects, man or vehicle, which can reduce massive manual work of surveillance video analysis.