Research On Obstacle Detection And Path Identification Method In Farmland Based On Binocular Vision And DSP

Agricultural mobile robot research is one of the hotpots in the robot research field.Inrelated research of agricultural mobile robot,obstacle detection and path identification are notonly the key of realizing robot to work, but also the core technology to achieve the realintelligent and autonomous move. Obstacle detection based on binocular stereo vision has theadvantage of wide range of detection and high accuracy, path identification based on DSPtechnology has the advantage of good real-time, fast processing speed,small size, low powerconsumption, and so on, so the research of this paper has important theoretical significanceand practical value.This paper focuses on the autonomous navigation of agricultural robots, developed arapid obstacle detection method in farmland based on binocular vision and path identificationmethod based on DSP technology, and setted up the software and hardware system, includingthe following work:(1) Established the imaging model of binocular vision, finished camera calibration withgroups of object images in different position, got the camera internal and external parametersby calculation methods and geometric transformations, after many experiments, the pixelerror of calibration was within0.2pixels, it can meet the requirement of the system and canbe used in subsequent calculations.(2) Introduced a variety of obstacles detection method in farmland based on binocularvision. technology, focus on the real-time inages acquisition, binocular camera calibration,binocular correction, stereo matching and depth measurement and so on, firstly the imagesreseived should be distorted by the result of camera calibration, and then used Bouguetalgorithm rectified the image pairs, then striked the disparity map using the local matchingalgorithm based on absolute difference, finally calculated the depth information of theobstacles by triangulation method. Programmed the code, and took a large number ofexperiments in the laboratory and field environment, we can achieve the detection of a varietyof obstacles, such as people, bricks, shovels, crater, etc. Experiments showed that when the detection range was less than2000mm, the errors were less,it showed that the method hasgood stability and robustness.(3) Introduced a path identification method based on DSP technology. Setted up thehardware experiment platform, it's core is the SEED-VPM642which is developed by BeijingHezhongda company, the hardware platform used a dedicated high-performance DSP as theprocessing core, using the idea of modular while the research process, including DSP system(CSL library functions, EMIFA interface, the EDMA, video codec chip, etc.)initialization,the input and output of video, the DSP core image processing algorithms, etc., completed thesimulation and testing of the entire DSP system. Experiments showed that the method had fastprocessing speed and better real-time, can meet the requirement of the system.