Research On Sequential Monte Carlo Based Hybrid Classifier Ensemble Tracking Method

Nowadays,visual tracking has become one of the most popular problems in computer vision.It can be widely used in Intelligent Surveillance,Vehicle Navigation,Human Computer Interaction,Military Defense and other fields.Although visual tracking has achieved many progresses,there is still a long way to realize adaptiveness.In many realistic tracking problems,the difficulties like the pose variation of target,abrupt motion,and complex background,always lead trackers to difficulty.In this paper,we made major research on the ensemble classifiers based object detection method.Ensemble tracking method treats object tracking as a binary classification problem.It locates the state of object by combining several weak classifiers into a strong classifier.This method becomes a hot issue in recent years due to its robustness and high accuracy.This paper made major research on the design of weak classifiers and the development of ensemble updating method.In detail: Firstly,the object region is represented as a patches pyramid,so as to realize robustness to target's global and local changes;Secondly,for the first time,the LDM classifier is introduced to the tracking problem,and is combined with SVM into a strong classifier for more accurate classification;Thirdly,in order to adapt with the complex variation of the object and its background,the ensemble parameters is regarded as sequential arriving states and estimated in a Bayesian manner.Specifically,a sequential Monte Carlo method is realized in order to realize the adaptiveness.In order to verify the validity of the proposed method,it is tested on a Benchmark data-base,including eighteen difficult video sequences.In addition,it is compared with five state-of-the-art tracking methods.The tracking results and comparison analysis on various complex videos demonstrated that,the proposed method performs well in handling many tracking difficulties such as complex background,object occlusion,illumination and so on.In conclusion,it is able to realize stability,high accuracy and better adaptability.