Monocular Vision-based Target Tracking Of Wheeled Mobile Robot

Vision-based wheeled mobile robots are widely used in military,medical,service and other industries because of their intelligent features such as high accuracy,visualization and strong maneuverability.These robots are generally integrated with high-performance hardware processing units,vision sensors and laser radars.Combining with sophisticated,advanced and intelligent algorithms,the robots can effectively track the moving targets in complex environments.However,the high costs make it difficult to promote consumer and family services.Therefore,many consumer service robots on the market use relatively low-cost assisted positioning sensors such as infrared and Bluetooth to realize simple tracking.Inevitably these robots have low tracking accuracy and short detecting distance as well as limited intelligent due to the lack of visual sensors.In order to solve the above contradictions,a low-cost monocular visual tracking wheeled mobile robot is designed in this paper.Moreover,visual servo control is implemented and visual tracking algorithm proposed to achieve continuous,accurate and rapid tracking.This dissertation uses one Arduino,one Raspberry Pi3 and one ordinary Logitech monocular camera to design a monocular visual vision robot under colsed loop PID vision control.By this end,the robot is costed down.In the aspect of visual algorithms and tracking algorithms,this dissertation first calibrates the camera and preprocess the images by filtering and histogram correction and so on to guarantee the accuracy of its post-vision algorithm and tracking algorithm.And then comparision of Mean Shift tracking algorithm and the tracking algorithm based on April Tag is done.It reveals that the accuracy of Mean Shift algorithm and anti-jamming performance are poor,while the tracking algorithm based on April Tag has higher precision and better anti-interference performance,but it need more computation resources.In this regard,three points are proposed in this dissertation to improve the tracking performance of the algorithm based on April Tag,so that it can be implemented in the low-cost hardware platform.The first one is that dynamic downsampling algorithm of the tracking algorithm based on the April Tag is theoretically proved by using the pinhole imaging camera model.Using the dynamic down-sampling algorithm,the robot can down sample more while the target is at a short distance to enbalance the visual accuracy and computation,and down sample less while the target is at a far distance to ensure the visual accuracy.Then figure out the function between the downsampled amplitude and visual sucessful tracking distance through the experimental datas,in order to dynamically choose a downsampled amplitude according to the distance of the target.In addition,the motion prediction is carried out based on the speed of the robot and the speed of the tracking target,so as to achieve an accurate distance of the target.The second one is to enhance the image so as to reduce the ambiguity and positioning inaccuracy due to downsampling.At last,secondary detection algorithm is designed in case of lossing the target occasionally,and motion planning algorithm is also designed to control the robot searching the target with special motion.After the tracking algorithm based on April Tag is optimized,the fastest processing speed up to 111 ms per frame,the positioning accuracy is less than 2 cm,the farthest precise recognition distance is more than 1.8 meters,and the quickly and smoothly tracking is achieved on the low-cost wheeled mobile robot.