Research On Lane Departure Warning System Based On Computer Vision

Research of Lane-Departure-Warning-System(LDWS) based on Computer Vision is now the focus of vehicle proactive safety system, and embedded real-time image-processing system is the hardware base of the LDWS. Aiming at the LDWS, the dissertation analized the image-processing system,and design an embedded system which can meet the real-time demands of the LDWS.Firstly, this paper proposes a Algorithms Classifing theory through analysing lots of vision algorithms. In this theory, vision algorithms are divided into several classes, so that, we can build proper computing architecture for each algorithm class.Then, the dissertation discusses the processing flow and the computing demands of the LDWS. Considering the real-time demands and reconfiguration, this dissertation introduces a hybrid architecture combining DSP computing and reconfigurable computing according to presented theory. Based on the presented architecture, the dissertation mainly discusses the data transportation problem, which has limited the CPU's dealing capability and dealing efficiency. In order to solve this problem, this paper builds a high performance data transportation bus, with which, computing notes(FPGA, DSP, SDRAM)are connected. In order to improve the efficiency of the data transportation among these notes, PDT(Peripheral Device Transportation) technology is introduced. Experiments show that the bandwidth of the data transportation bus with PDT can reach 0.77GBytes/s, which can meet the demands of the LDWS.In the end, this paper performs a complete vision computing task on the proposed architecture, and the result indicates that system can meet the real-time demands. Reconfiguration and general purpose are also considered in the architecture design.