Object Tracking Algorithm Based On Vision And Its Application On Mobile Robot

The rapid development of computer vision makes the visual target tracking technology to beone of the leading topics. Significant progress has been made in targets tracking by lots ofresearchers in the last few years. The technology has been applied to video surveillance, robotlocation and tracking, multi-robot formation and lunar exploration, ect., with wide applicationprospect. However, the existing tracking algorithms can not guarantee the tracking effect in thecomplex environment, such as fast illumination variation or object occlusion. So a epipolarconstraint principle, as well as a SIFT algorithm based on it is proposed in this paper.In this paper, a detailed analysis of SIFT algorithm is presented, and an improved SIFTmatching algorithm is proposed based on SIFT algorithm. The matching accuracy of classical SIFTalgorithm will significantly decline when amplifying the matching threshold. So the epipolarconstraint principle, as well as a SIFT algorithm based on it is proposed. In the approach, theregular SIFT method is used to extract distinctive invariant features from two images, and the initialmatches of images can be obtained by the SIFT algorithm with a low matching threshold. Then thefundamental matrix of the epipolar constraint is calculated by the RANSAC algorithm. The epipolarconstraint method is followed to eliminate the false matches from the coarse ones which areobtained by the SIFT algorithm with a high matching threshold. The experimental results show thatthe approach can improve the matching accuracy effectively and more feature matches can beobtained.Although the target tracking algorithm based on color features is very popular ,it is unstable. Atracking algorithm with epipolar constraint based SIFT algorithm and particle filter is proposed. Thealgorithm is establishes the target template with the feature vectors of the SIFT feature point, andestablishes the candidate template with SIFT feature vectors, using a particle filter method.Choosing likelihood function to measure the similarity between target template and candidatetemplates, the proposed algorithm estimates the target state, to perform an effective target tracing.Experiments show that the proposed algorithm can solve the problem of tracking instability whenthe background color is similar to target, but also has strong robustness when target’s posture orshape changes.In this paper,the target tracking algorithm based on epipolar constraint of SIFT feature andparticle filter is applied in the robot of Subsumption behavior architecture to design motion controlscheme. The mission of tracking is performed, it can receive Target’s two dimensional position anddepth information are acquired by the nDepth binocular vision sensor ,which is produced by Focus Robotics Co.,Ltd, to over that the monocular vision can not get depth information. The resultsillustrate that the mobile robot with binocular vision can effectively complete the task of movingtarget tracking.