Robot Visual Feedback System Base On Deep Reinforcement Learning

At present,with the development of the medical industry in the direction of intelligence and automation,surgical navigators have received widespread attention from medical workers.Aiming at the problems of high economic cost and complicated implementation process of surgical navigator,this paper builds a visual system of surgical navigator based on deep learning and reinforcement learning.The visual feedback system can not only detect multiple surgical auxiliary targets,but also perform three-dimensional positioning of the surgical auxiliary targets.First of all,in order to solve the problem of the need for manual assistance in the detection of the center of the image of the surgical auxiliary target,this paper uses the YOLOv3 algorithm to determine the category of the surgical auxiliary target and estimate the position,while reducing the interference of the external environment.In addition,this paper combines the Canny algorithm and the HSV color model to preprocess the picture,and uses the Hough center detection algorithm to determine the coordinates of the center pixel of the surgically assisted target,and is used as a foundation for determining the three-dimensional coordinates.Secondly,in view of the problem of more redundant areas in the YOLOv3 algorithm,this paper introduces the DQN algorithm to detect surgically assisted targets.By modeling and training the DQN algorithm,the model can decide the action selection based on the current features,reducing the number of candidate regions in the target detection task.After that,the research on the three-dimensional positioning of the surgical auxiliary target determined by the monocular camera was carried out.First,the internal parameters of the monocular camera are determined by Zhang’s calibration algorithm.Then,the parallax algorithm is used to calculate the three-dimensional coordinates of the surgical assist target in the camera coordinate system.Using the conversion relationship between the camera coordinate system and the robot arm coordinate system,the three-dimensional coordinates of the center of the surgical assist target in the robot arm coordinate system are obtained.In addition,a three-dimensional coordinate error model of the surgical auxiliary target is established,and improvement measures to improve the accuracy of the experiment are proposed.Finally,this paper designs and selects the hardware system of surgical navigator.At the same time,in order to obtain the output results of the algorithm intuitively,this paper completes the programming of the upper computer interface.In addition,this article compares and analyzes the experimental results and proposes corresponding improvement measures.