A Study On Moving Object Detection Based On ADSP-BF533 And Its Application In Car-carrying Intelligent System

This paper focused on moving pedestrian detection and its hardware realization. The works of this paper are as followed:In the scenario of moving objects detection and retrieval under a static background, an improved algorithm based on multi-frame difference was presented. A dynamic value was used to threshold the difference results for binary image in the improved algorithm. According to the components characteristics of retrieved targets, multi-frame difference can provide more information than the two-frame difference. As a result, the improved algorithm retrieved moving objects more accurately than the original one.In the scenario of moving objects detection and retrieval under a dynamic background, an improved method was presented based on gray projection algorithm. To reduce the computational consumption and expand the range of global motion estimation, an adaptive dynamic range of global motion estimation was used in the improved algorithm.The system hardware based on the digital signal processor ADSP-BF533 was designed and implemented. The system design includes the video CODEC, the system output control interface and the policy design of instruction and data cache. In the video CODEC module, the design based on ADV7183 was proposed. To improve the data access efficiency, a data structure called ping-pong buffer was presented for the input video data. In the policy design of instruction and data cache module, the instruction cache and data cache were applied to the system design and cooperated with each other effectively. This improves the program performance and the data access efficiency.Combining with the advantage of multiplication and addition operation by processor ADSP-BF533, the system achieves good experimental results and meets the requirement of real-time under the translating background, which is applicable for car-carrying intelligent system.