Visual Tracking Algorithm Based On Feature Extraction And Color Model

The research of visual tracking based on video image sequence is animportant area of computer vision research, which has been widely appliedin fields such as video monitoring system, intelligent transportation system,national defense and military, human-machine interaction, etc.This thesis studies and investigates the current major visual trackingalgorithms, and proposes the visual tracking algorithm based on featureextraction and color model on the foundation of ORB feature descriptionalgorithm and classical CAMShift object tracking algorithm. Firstly in theforeground extraction step, an improved three frame difference method isproposed based on Sobel edge detection operator and basic morphologicalprocessing, which improves the performance of moving object extraction.Then do research into the ORB feature description algorithm andintroduces it into the process of visual tracking by building and upgradingthe feature template of target, and through comparison experiment amongSIFT and SURF, verifies the efficiency of ORB feature using in thereal-time tracking.Moreover, the CAMShift algorithm has been introduced in detail,including the basic principle and the merits. An adaptive color thresholdsegmentation algorithm is proposed to improve the accuracy of the target’scolor model. To solve the defects of CAMShift, the target motionforecasting module based on Kalman filter has been brought into theprocedure of CAMShift, which has been verified to be effective byexperiment in the scenario when target is under occlusion. Afterwards inorder to improve the performance of high-speed tracking, a target missingand relocation module has been added to the improved CAMShift, and theORB feature information is used to revise the search window of CAMShift.Eventually, by quantitative analysis the result of two experiments underdifferent scenarios, validate the new algorithm has higher veracity andreal-time quality. In addition, a real-time visual tracking GUI system isdeveloped by using MFC and OpenCV, and overall framework andfunction module of system is designed and implemented, which improves the efficiency of visual tracking experiment and the practicability of thenew tracking algorithm.