| Visual tracking is always being an active research topic in Information technology, becauseof its wide application in the automatic object recognition, intelligent surveillance, vehiclenavigation, vehicle navigation, human computer interaction, and other disciplines. Photoelectricimaging visual tracking for over-ground surveillance and tracking, used into target automatictracking and ranging in the vehicular photoelectric platform.This thesis focuses on visual tracking algorithm based on DSP. To overcome the presences ofnoise，occlusion, background clutter and illumination changes in visual tracking algorithm, it isneed to develop a robust visual tracking framework. The particle filter framework built in thiswork casts the tracking problem as finding sparse representation of candidate target. And then,proposing a solution to settle up the sample impoverishment problem brought from theresampling method in the particle filter framework. Here are researching contents in this work:Bring up the genetic bounded resampling algorithm. Under the condition of no sacrificingthe resampling accuracy, use the bounded samples resampling to reduce the time-consumingcalculation of the sampling values. Then, use genetic algorithms to increase the diversity of thesample in order to reduce sample degradation problem, brought from resampling in the particlefilter framework.Detect occlusion. Using dynamic template to capture surface changes, to stop the unsuitedtracking results from adding into the template set, it detected the occlusion before the resultsadded into the set. And then control the updating of templates.Display the tracking results visually. Build the DSP platform, using the CCSLink tools tocombine the DSP and MATLAB in order to perform the data transmission. The DSP chip calltracking algorithm and return the results into MATLAB workspace. Then the tracking results areturned into intuitive picture format through the capability of powerful data processing inMATLAB. |
PDF Full Text Request