Research On Binocular Visual Impairment Perception Based On Yolov5

Agricultural robots are becoming the research focus of researchers and related enterprises with its automation,high intelligence,high efficiency and other operational advantages.With the deepening of research and increasing market demand,agricultural robots and agricultural intelligent equipment will become an important development direction to enhance the core competitiveness of related enterprises,drive scientific and technological innovation and promote the improvement of agricultural production efficiency.Machine tiller is a kind of paddy field power machinery,equipped with corresponding farm tools,can complete the plowing,rotary tillage,leveling and harvesting operations.Its simple structure,low cost,and operation efficiency and economic efficiency is relatively high,deeply loved by the majority of farmers.With the transfer of rural labor force to the city,the traditional labor is gradually replaced by the advanced labor production mode.The manufacturer of the machine tiller boat is considering how to upgrade the machine tiller boat with high intelligence and high efficiency,so as to realize unmanned driving.The longitude and latitude coordinate values of the machine tiller can be calculated in real time by GPS and combined with path planning.However,in paddy field environment,there are often agricultural vehicles,pedestrians and other obstacles,so how to identify obstacles in the working scene and avoid them has become the key research content.In this paper,the recognition of obstacles and the depth detection of obstacles are studied and analyzed,and the camera calibration,stereo correction and stereo matching in binocular vision are studied.Mainly completed the following work:(1)The application background and current research significance of the machine-tiller ship are introduced.The research status of obstacle recognition and stereo matching algorithm based on binocular vision is reviewed at home and abroad.Finally,the technical route adopted in this study is explained.(2)The calibration and stereopole correction technology of binocular camera are studied,and the camera imaging principle and four coordinate systems in the camera imaging model are introduced.By using Zhang’s calibration method,the calibration of binocular camera and stereo image correction are completed,and the specific parameters obtained are verified.(3)The convolutional neural network process is introduced and various first-order and second-order neural networks are briefly introduced,among which the YOLOv5 network used in this paper is emphatically studied.I created my own agricultural obstacle data set and applied it in YOLOv5 network for detection and comparison with other networks to verify the advantages of YOLOv5.In this chapter,I also studied stereo matching,built stereo visual ranging system,conducted stereo matching experiments,and analyzed the experimental results,which met the requirements of the scheme.(4)Designed and manufactured a self-developed hydraulic four-wheel-drive agricultural vehicle,and conducted preliminary walking experiments on the vehicle.Explored a preliminary plan to combine the agricultural vehicle with visual obstacle perception,in order to adjust the agricultural vehicle’s direction and speed in real-time,thus improving its driving precision and operational efficiency.