Research On Target Tracking Algorithm Under Complex Environment And Its Application In Mobile Robot

Target tracking technology has been widely used in many fields,such as military, industrial, medical and security,leading it to be a research hot spot. However, because of complexity of the background ,self-posture, illumination variation, structural deformation ,object occlusion, etc.,it is difficult to tracking object both robustly and computational efficiently. This paper focuses on the research of both robustness and real-time target tracking algorithm.The major work of the paper are as follows:1.In terms of description of target, several features are combined to describe the target. An optimized feature extraction algorithm,which can significantly reduce comptutation of feature , is proposed in the process of template updating.This method is helpful for the purpose of distinguishing the target from the complex background.2.In this paper, target detection algorithm has been studied,and then an improved AdaBoost (Adaptive Boosting) algorithm,based on genetic algorithm and correlation between classifiers,was proposed. Experiments showed that the improved Adaboost algorithm has higher recognition accuracy than traditional artificial neural network in the same sample set,and faster detection speed than that of support vector machine.3.In respect to target tracking algorithm, an improved particle filter tracking algorithm has been presented. In this algorithm ,observed value was used to optimize the proposal distribution, leading to dramatic reduction of particles and effective improvement of the efficiency of particle filter sampling.At the same time, a fast template update strategy was presented,owing to which the template can be quickly updated and the observation values can be fastly received.The predicted value by particle filter was used to limit target detection range,which can enhance detecting speed. Experiments showed that , compared to traditional particle filter, this improved algorithm has significantly reduced the amount of particles needed by target tracking and performs better than camshift algorithm. At last, this paper's algorithm is applied to mobile robot platform,and to actively track first target in vision.