Design And Implementation Of Obstacle Detection System For Lawn Mower

Compared with the traditional mower,the modern unmanned intelligent mower has the advantages of high efficiency,safety and reliability.However,due to the complex outdoor lawn environment,such as human and animal walking,tree shelter and so on,there are security,real-time and frequent false alarm problems in the operation process of intelligent mower.Therefore,on the premise of guaranteeing the effect of multi-obstacle detection,improving the speed of obstacle detection is the key to realize engineering application and improve the level of intelligence of intelligent mower.At the same time,due to the need for intelligent mower to avoid movable obstacles,there may be a situation of lawn omission.Meanwhile,in order to ensure the comprehensive coverage of the mower operation area,and ultimately to improve the efficiency of mower operation and the overall coverage of the mower operation area,it is necessary to design the function of the mower based on the efficient obstacle detection system.This paper mainly relies on the school-enterprise cooperation project "Research on New Technology of Intelligent Recognition of Lawn Obstacles by Mowing Robot".Through the project,the design and implementation of obstacle detection system for the moving end of the mower are carried out.The main contents are as follows:Aiming at the problem of low efficiency of obstacle detection caused by complex lawn environment,a set of high efficiency obstacle detection scheme which is mounted on the moving end of lawn mower is presented.Firstly,YOLOV3 deep learning model which has good detection effect and strong expansibility is selected as the basic model of obstacle detection.The model is used to identify common obstacles on lawn,including people,dogs and lawn signs.Then,through the combination of models and the acceleration of the two steps on the embedded end,the speed of model forward reasoning is improved.By combining the feature extraction network darknet-19 of YOLOV2 with the multi-level prediction network structure of YOLOV3,the Logsitic loss function proposed in YOLOV3 is adopted.These methods not only satisfy the needs of the project,but also reduce the complexity of the model as much as possible and improve the operation speed of the model.Finally,the combined model is processed in single-precision mode by using TensorRt scientific operation tool,which improves the forward reasoning speed of the compressed model transplanted to NIVIDIA Jetson TX2,and achieves efficient obstacle detection at the moving end of the mower.In order to solve the problems of low efficiency and missing mowing caused by movable obstacles in lawn,this paper integrates pedestrian hinting function and geographic location information recording function on the obstacle detection system of lawn mower.The function of recording geographic position mainly relies on ATGM336H-5N multi-mode positioning module and Aduino sensor to analyze satellite signal,and transmit data to NIVIDIA Jetson TX2 control center module to realize the acquisition of mower geographic position information.The above functions are realized by integrating software and hardware.These functions can reduce the number of times the mower avoids dynamic obstacles and prevent the lawn from missing mowing,thus filling in the blank of the mower obstacle detection system.In this paper,by making mower obstacle data sets,compressing and training in-depth learning target detection model,accelerating embedded model and integrating software and hardware of positioning module,a mower obstacle detection system which integrates multiple obstacle detection,pedestrian hints and recording mower geographic location information functions is realize.This research lays a solid foundation for the engineering application of intelligent mower.