Based On Real-time Image Processing For Visual Navigation And Fpga

Autonomous Land Vehicle (ALV) is an intelligent mobile robot, which can run autonomously and continuously on road or cross-country. Its investigation, which involves theories and technologies of many disciplines, is of great value in research and application. The visual perception module is always one of the most important modules of ALV. Load detection and stereo vision are two key techniques of the visual perception module. This thesis aims at the study and design of the vision navigation system for the ALV.Accurate and real time capture of road image is the foundation of visual sense research, and firstly this thesis does some research to this problem. The second chapter of thesis has designed the dedicated purposed video capture card described the complete hardware implementation. This scheme uses FPGA as the core chip to control the image collection and data bus transfers. The video decoding chip SAA7115 of Philips corporation and the PCI interface chip PCI9054 are used. It is proved by experiments that this video capture card can get real-time road image precisely.Image filtering and edge detection is one of the key procedures in complex vision processing such as road edge and obstacle detection in vision-based navigation systems. Its performance and processing time directly affects the performance of the subsequent image processing and the total response time in vision-based navigation systems. The third chapter introduces several image filtering and edge detection methods that used to road detection in vision-based navigation systems, and analyses the implementation structure of Sobel operator, LoG template and Canny operator. Takes the characteristic of FPGA into consideration, proposes a method of image template filtering realization based on distributed arithmetic. The Canny operator edge detection method is also optimized and implemented in a more efficient way.At last introduces the stereo match algorithm, and the preprocessing and geometry emendation processing before stereo match stage. Several different similar measurements of area-based stereo matching are discussed, and the effects are compared. The size of matching window and the accelerating techniques are studied, and the implementation structure of area based stereo matching is described.