Research On Pedestrian Detection And Obstacle Avoidance Path Planning Of Unmanned Spray Robot Using Binocular Vision Method

The planting area of economic crops in orchard is large.Traditional way of using artificial orchard to spray is low efficiency,high labor intensity and easy poisoning.In order to improve efficiency of orchard spraying,reduce labor cost and reduce harm of pesticides to the workers,unmanned operation sprayer should be developed.In order to ensure safety of the unmanned operation of sprayers,reliable obstacle detection and obstacle avoidance functions must be provided.In order to realize autonomous spraying operation of unmanned sprayer in orchard,this paper studies the detection and obstacle avoidance control of pedestrians in orchard,so as to ensure no personal safety accidents.The main research contents are as follows:(1)The control system of the unmanned sprayer is designed.The system consists of data acquisition and processing system and motion control system.Data acquisition and processing system uses a binocular vision sensor to obtain the image information of working scene.Embedded AI computer processes the data to obtain spatial location information of obstacles and completes obstacle avoidance path planning task.Motion control system acquires control information of upper computer through master processor,and acquires information of hardware sensor in real time.The control command is sent to driver at the bottom to control the driving of sprayer.(2)Improved SSD(single shot multibox detector)object detection algorithm based on deep learning is proposed and combines with binocular vision system to detect the spatial position of pedestrian obstacles.Lightweight network mobilenetv2 is used as basic layer network of SSD object detection algorithm to reduce time and computation in the process of extracting image features.The auxiliary layer network uses the improved inverse residual block as basic structure for position prediction,which avoids information loss caused by down sampling operation while integrating multiple scales,The experimental results show that the average accuracy and recall rate of the improved SSD target detection algorithm are 97.46% and 91.65%,respectively,which are higher than the original model’s 96.87% and 88.51%,and detection speed is 62.50 frames/s.Average relative error of positioning accuracy of pedestrian obstacles in Z direction is 1.64%,and the average relative error of X direction is 1.83%.This method can better detect and locate pedestrians in orchard environment,and provide a basis for avoiding obstacles of unmanned sprayers.(3)Obstacle avoidance path planning for unmanned sprayer under orchard environment is studied.First,orchards environment model is established according to the known prior information of orchard environment,and then the path planning for the full coverage of fruit trees is completed.Next,the defects of traditional artificial potential field method are improved,and the improved artificial potential field method is used to avoid the local pedestrian obstacles which are not considered in global path planning,and experiment is carried out in MATLAB and in outdoor to verify the effectiveness of the algorithm.The experimental results show that improved artificial potential field method can effectively complete obstacle avoidance path planning task and meet the real-time requirements.