Research And Implementation Of Visual Analysis Algorithm Based On Da Vanci Platform

The intelligent vision analysis technology, one of the key research directions in the computer vision field, aims to detect, track and recognize moving objects in the monitoring scenes. The intelligent vision system that can be equipped to various places with large-scale is strongly needed. And conventional PC based video surveillance system is out of date. Therefore, it is of theoretical meaning and practical value to develop embedded intelligent video system to adapt the variety surveillance occasions with large-scale.Pedestrians are the main objects for the surveillance system. Therefore, research on the detection, tracking and recognition of pedestrians remains the core issue and major direction in computer vision domain. How to develop human motion analysis algorithm and transplant it to the embedded platform is the development bottleneck of the intelligent video system.In chapter 1, the significance of this thesis is presented together with a brief summary of the present research status.In chapter 2, a statistic principle based detection method is proposed with the unit region background model. Firstly, the original video data is compressed via index image method. And then the the efficiency of detection has been enhanced greatly via the unit region statistic method and background partly updated strategy. Meanwhile, shadow suppression and separation of two linked objects are processed by the foreground pixel group histogram modify method. The effective and efficient detection algorithm lays foundation for the realization of embedded intelligent vision system. Then the features fusion based tracking method is applied to enhance the tracking accuracy with twice feature matching method.In chapter 3, Haar feature based Ada-Boost algorithm is applied for pedestrian recognition. Firstly richer sample features are abstracted via spare granular constructed Haar features. The weak classifiers are trained off line and then cascad to compose a strong classifier for the effective recognition and low-cost computation.Chapter 4 optimized the program of the intelligent vision analysis algorithm. And the VICP is used for the integral image computation to reduce the burden of CPU. Finally, the program is packed to a standard module to make it independent to hardware platform and callable by Linux application.The final chapter concludes the achievements of the whole research and prospect of the future research.