The Research On Key Problems Of Human Motion Target Detection And Tracking

Human motion analysis is an important technology of computer vision and biological combination. It has widely application in many fields,such as military defense, video surveillance,human-computer interaction, medical diagnosis and commercial. At present, domestic and foreign scholars have done a lot of research on this field, and achieved some results. But because of the complexity of human shape and motion, human motion analysis is still facing many problems in the theory and application, This paper is based on the previous work, Further study the key problems in the process of human motion detection:background modeling and moving object extraction, human motion target automatic tracking process:template update and trajectory prediction.In the aspect of human moving target detection,Through the comparative and analysis of various human detection algorithm, Using edge detection, gray correlation calculation and image block processing technique to improve the surrendra background modeling method. In the target extraction process, Selection the adaptive iterative threshold method, morphological processing, connectivity analysis to optimize the extraction techniques results in the process of selection. The experimental results show that the improved algorithm deals with adjacent pixels as a whole,eliminating the error caused by a single pixel perturbation, effectively overco-me the influence of illumination change, with real-time and high accuracy.In the aspect of human motion tracking, Focus on the analysis of the principle of Mean-shift algorithm and its deficiencies in the human motion tracking, We improve it: realizing automatic tracking by using human target detection information to initialize the Mean-shift parameters; Through introduction combine of the least square linear prediction and curve prediction method effectively solving the positioning error and target occlusion problem;By using the template weighted method to improve the stability Mean-shift algorithm; Finally, according to the centroid position of tracking to record human target trajectory.The experimental results show that this algorithm realizes human target automation tracking, enhancing the Mean-shift algorithm tracking effectiveness in target pose variations and illumination changes, overcoming the target is disrupting by interference or occlusion, assuring the reliability of matching, having obvious advantages in stability and real-time.