Research Of Intelligent Traffic Video Detector Based On DaVinci

Video detection technology is not a new concept in the transport sector, as there have been much relatively research overseas. However, the technology is still in a relatively backward stage interiorly, and the practical application is still a long way to go. There are many theoretical algorithms about video detection, but how to choose an applied platform that can support the algorithms, how to develop some algorithms to deal with the practical application are very meaningful. And that is my major work.This paper is mainly to research the transplantation and optimization of traffic video detection algorithms on TMS320DM6446. And how to build a embedded systems that meet the requirements of the vehicle detection by Davinci platform.It describe several different methods to detect transport and analysis the superiority of video processing briefly in the paper. Also we have systematically studied the features and advantages of Davinci platform, detail the overall design of the video detector. Introduce of the technology of the program is divided into two parts, hardware and software.The key technologies are algorithms'transplantation and optimization on DSP, as well as the embedded applications. The subject needs cooperation, and detection algorithm is the core technology, but it is not the emphasis of this paper. We are concerned about the forementioned transplantation, optimization and embedded system design. The difficult point of transplantation is mainly embodied in dual-core processing, in other words, how to send the data on ARM to DSP, and how to send DSP results to ARM. The focus of optimization is how to effectively use the software and hardware resources of Davinci to improve the efficiency of the entire system. Systematic design of embedded application is another focus of this article, which is a heavy work and covers a wide range.We have tested the Optimized embedded systems detailedly and comprehensively. The ARM CPU occupancy rate has decreased to 40%, and the DSP CPU occupancy rate decreased to 30%, which fully satisfy the real-time requirements. And, this makes improvement of algorithms possible in future.