Moving Target Detection And Tracking Of The Key Problems In The Research And Implementation

Computer vision is one of today’s most active areas for research. It’s thetransformation of data from a still or video camera into either a decision or a newrepresentation. As a mature cross-discipline, computer vision is closely related with thefield of computer graphics, artificial intelligence, artificial neural networks and so on.Its rapid development makes it possible for practical application in various fields,which includes product testing, navigation of mobile robot, dealing with medicalimage and the intelligent monitoring.Our paper focus on the detection and tracking of moving targets in video, and thekey technologies are studied, analyzed and realized based on the OpenCV library. Ourmain tasks are list below.In the field of moving target detection, our paper describes and compares the threemain algorithms and their variants firstly. After fully considered the advantages anddisadvantages, the method of background subtraction has been chosen and the mostwelcome modes for background are introduced, such as Gaussian model and Codebookmode. Based on Codebook model, a novel model was proposed and realized to makethe detection method more closely to practical. The results show that the proposedmodel can achieve satisfactory results in both accuracy and real-time.In the field of tracking moving targets, the methods of de–noising and how toextraction of connected domains has been described firstly. After that, two mainpredictors are described: Kalman and Condensation. Kalman filter can give lessaccuracy but demand much less computation, and has been chosen for the prediction oftargets’ positions and velocities. At last, whether the targets in two adjacent framebelongs the same subject or not can be told according to its color and shape’s similarity.The results also showed that our tracking method is effective.