Research On The Following System Of Tracked Vehicle Based On Vision

In recent years,with the development of science and technology,robot technology has been increasingly applied in the field of agriculture.Agricultural robots can not only improve work efficiency but also enhance work quality and solve the problem of labor shortage.For example,the use of mobile mechanical equipment such as tracked vehicles is very common in agricultural production.However,operating these machines requires a lot of manpower and time,and the difficulty in operating them increases particularly in large-scale orchard.To solve these problems,this study designs a vision-based tracked vehicle following system for orchard tracked vehicles.The specific steps are as follows:(1)Develop a following system using the ROS platform.Based on the scenario requirements,a test vehicle is selected,and an orchard tracked transport vehicle is designed.Control nodes for the two tracked chassis(orchard transport tracked vehicle chassis,test vehicle chassis)are written separately.To meet the vehicle’ s functional needs,nodes are written.The communication between nodes is implemented through the topic communication method.(2)Design the target detection module.According to the requirements of the following target and the actual orchard environment,two detection methods are designed to complete the detection of the following target.The first method is to attach a specific information AR code to the target that needs to be followed,and indirectly obtain the position information of the following target through the AR code.The detection based on the AR code mainly tracks the posture information of the AR code in real-time by installing an AR tag tracking library.The second method mainly detects the following target directly based on the YOLOv5 s network,in which data is collected for target detection in indoor,orchard,and occlusion scenarios.The target to be detected is marked using a sprite annotation assistant.The training and validation datasets are separately created.To address the problem of different farthest detection distances and detection speeds for different sizes of AR codes,the farthest distance test for AR codes was carried out under three lighting conditions.The test results show that the farthest recognition distance for a 15cm*15cm AR code in a horizontal straight line during following based on an AR code is 5.25 m,6.05 m when the angle with the horizontal is 30 degrees,and2.78 m when the angle with the horizontal is 60 degrees,which meets the following requirements of the application scenario.Considering the possible differences in the detection speed and accuracy of different parts of the body,the detection speed and accuracy of the upper body,legs,and the whole body in the orchard environment are compared using the validation set.The comparison results show that detecting the entire body has a higher accuracy than detecting certain body parts,and it only takes 105 ms to detect one image.(3)Design Following Control Module.Based on the 3D positional information derived from the results of the AR code detection,a strategy was designed to control the tracked vehicle to follow the target at a certain distance.It is based on the 2D pixel coordinate information determined by YOLOv5 s object detection.To address the possibility of target loss during the following process,a re-detection and following strategy was designed.(4)Test tracked vehicle following system.After building the entire following system,the test vehicle was tested by AR code following testing experiments and orchard environment following performance experiments.The former test results show the process of following in a straight line has a good following effect,but the detection effect is poor at the corner of the orchard due to the loss of AR code in the field of view.When the turning radius is 1.5m based on the YOLOv5 s target detection result following,the tracked vehicle has a good following effect.The latter test showed that when a person moved at a speed of 1.5 m/s and the tracked vehicle followed at a speed of 1 m/s,the test vehicle would lose track of the person after 12 seconds of following.During the process of following the target in a straight line,the narrowest following width of the test vehicle is1 m.When the target is lost,under normal lighting conditions and when the person’s moving speed is below 1m/s,the test vehicle can search for and follow the target again.During real-time following,the system can detect the target in real-time and output the position information,test vehicle speed,and turning angle information of the detected target,and can maintain a certain distance to follow the target.After testing,the system was transplanted onto the orchard tracked transport vehicle in the orchard,and following tests of the orchard tracked vehicle were conducted.It has a good following effect when the person’s moving speed is below the fixed speed of the orchard tracked vehicle at 1m/s in the orchard environment.