Research On Technology Of Tracking, Detection And Recognition For Hand Gesture And System Implementation

Hand gesture recognition technology based on the computer vision is animportant research field of computer vision. In particular, monocular ordinary cameragesture recognition is one of the most challenging problems in computer visionresearch. This paper focuses on the monocular and ordinary camera isolated gesturerecognition and presents a monocular complex background gesture recognitionresearch and development. The exploration and study about affectting systemrobustness and real-time factors be carried out and research is carried out in thefollowing arear:1. Improvement of adaptive tracer, target tracked is a set of points. There is aphenomenon that some of tracked points may be lost This paper put forward amethod that forward-backward error and similarity asses the forecast quality. Itsuccessful achieve that the tracker adaptive to the target and extent the time of righttracking target.2. Research on recovery mechanism of tracker. For tracker has no recoverymechanism after tracking failure. Tracking-by-detection method is proposed.Thetracker runs in parallel with detector and re-initialization after its failure. For the typeof real–time detector based on a scanning window strategy: the input image isscanned across positions and scales,at each sub-window classifier decides aboutpresent of the object,the classifier is used high frequently,so there are strictrequirements on the performance of the classifier. Random forest classifier isproposed. The training sample datebase is modeled through “growing events andpruning events and the classifier online learns on the database.3. For discovering the start point of the meaningful gesture is difficult.This paper proposes a mouse control technology that the start point and end point of theeffective trajectory is setted through mouse events,and the problem is solved.For separate features can not achieve high recognition rate.The best combination offeatures (position, angle, speed) is extracted and achieve the improvement ofrecognition rate.Base on the above-mentioned algorithm research, a simple gesture recognitionapplication system is developed. This system can identify gesture symbols (0-9) and(A-Z), and finally through experiments the result show that the system has a highrecognition rate.