Application Research Based On Multi Scale Space Feature Extraction-Visual Tracking

Computer vision is an important research field in the21st century, which combines thelatest technology, such as image processing,pattern recognition, artificial intelligence and soon. Computer vision technology has broad applications prospect in robot vision navigation、intelligent transportation、biomedical、video monitoring、human-computer interaction militaryand other fields. Target tracking technology is an important branch of computer vision, whichis an interesting topic among researchers. Target tracking task is to seek to the position in thevideo frame, and than analyzing the behavior of the target. A large number of trackingalgorithms have been proposed, but the situation in the real world is very complex. Trackingalgorithm may affected by the light change、posture、 movement. The current trackingalgorithm is not mature, and researching the robustness tracking algorithm is still to be achallenging task.After researching the successful experience of predecessors, I do the following work:First,making deep research in scale invariant feature transform (SIFT algorithm).Inorder to solve the real-time and robust in SIFT algorithm,A new feature extractionalgorithm(MSRFE) is proposed,which is based on multi scale space robust feature extraction.A new Gauss template is built in the MSRFE algorithm, and the big step template is rejected.The bidirectional matching is used in the matching step, and changing the measure of similarjudge. The improved algorithm has good robust and real-time.Second,after studied the traditional tracking algorithm and recent trackingalgorithm,there new tracking algorithms were proposed in this paper.firstly,multi scale spacerobust feature matching algorithm is used to track the small target in the video. Extracting thefeatures in the former frame， and extracting the features in the rear frame,where theextracting size is three times than the former frame target area.Matching the features betweenformer frame and rear frame,and calculating the center points of the matching results,whichis regarded as the center of the target;Secondly, a new particle filter tracking algorithm isproposed,which is based on multi-scale space robust feature matching. Using the samematching algorithm as the first matching algorithm to accomplish the matching step,andcalculating the center the points among the matching points. Particles is distributed by theconditions of the center point in the next frame，and we can overcome the degradation ofparticles in this way,and we can achieve the robust tacking;Thirdly,a new compression sensetracking algorithm is proposed,which is based on the multi scale space L-HAAR featureextraction in the DOG space,CT tracking algorithm is sensitive to the light and shutterchange. In order to solve the problem, the features were extract from the DOG space，whichis called DOG L-HAAR features, The feature is very robust to the light and shutter change,and we can achieve the robust tacking by the new CT tracking algorithm.Finally,after doing the experiment, analyzing the advantages and disadvantages of thetracking algorithms.After giving the analysis of the tracking precision and real-time.thesuperiority of three new tracking algorithms can be present.