Object Detection And Tracking Techniques Based On DM6446

Visual object detection and tracking technology has an important strategic sense inmilitary field and has great practical value in civil field, so that it becomes the popular topicboth at home and abroad for the numerous universities, institutes and related enterprises. Withthe development of embedded technology, the visual object detection and tracking based onembedded platform has a very broad application prospects.In visual object detection, in order to overcome the faultiness of the traditionalbackground subtraction and temporal difference, an improved object detection methodcombined the edges and the symmetrical temporal difference to dynamically construct thebackground model, then the object information is detected by background subtraction. Theexperimental results show that this method has adaptability of dynamic background, goodcapability of object feature extraction, robust character of the scene noise, detection effect isbetter than that background subtraction and temporal difference.In visual tracking, this thesis first researched the particle filter visual tracking and themean shift visual tracking. Then according to the characteristic of the two visual trackingmethods, an improved method based on the particle filter tracking was proposed, aiming toimprove the particle degradation and particle impoverishment problem. In this method, wemade two improvements about particle sampling and re-sampling. On the one hand, weproposed adaptively incorporating mean shift into a particle filter to optimize the particlesampling process. On the other hand, we introduced the genetic evolution strategy improvingre-sampling process to increase the diversity of particles. This improved methods is both goodreal-time and robustness.In the application of Embedded Technology, we designed a embedded visual trackingsystem based on DM6446. We designs both the overall framework of this system and its eachmodule, then we build the system developing environment and make programmingoptimization. Finally, we performed visual tracking experiments on this embedded visualtracking system. The experiment results show that the visual tracking algorithm we proposedis robust, the embedded platform can meet the real-time requirement.