An Object Tracking Method Based On Classification And Recognition

The actual needs in practical application promote the development of manydisciplines. As one of the representatives of intelligent technology, computer vision isfacing unprecedented opportunities and challenges. Computer vision research has attractedmany top scientists, engineers invented and improved many computer vision products, andtake them to application in Satellite navigation, industrial inspection, security monitoring,medical imaging and other industries, and gain a great success.Target tracking is a basic research in the field of computer vision research, also a taskthat has been studied for a long time without much breakthrough so far. Experts proposedvarious kinds of theories or methods to solve this problem. Some of these methods arecombined with specific practical applications and gain good performance in these fields.some new theories are proposed to provide new ideas for future research. There areseveral major problem which has not been resolved in target tracking, such as how toselect the best appearance model and features of target, how to achieve long-term stabilityof the tracking, how to achieve the same speed as the human eye, how to identifysuspected false targets, how to adapt to the appearance changes of target, how to predictthe state information of the target when it’s hidden.Based on in-depth analysis these existing problems, this study give some preliminarysolution ideas. First random fern classification method is proposed to quickly filterbackground area, with the use of local binary image feature which is very simple andefficient, we are able to quickly filter out the background area with low similarity to target.For those objects with a relatively high degree of similarity of the target, we studied amethod based on the background context information restraint, and reduced the falsealarm rate. Finally, we propose a method based on improved incremental principalcomponent analysis, which can deal with sparse random noise, finally improved target tracking stability.