| The endoscopic instrument tracking algorithm is one of the core algorithms of the robotic arm automatic navigation system and one of the core algorithms of the medical surgery robot.At present,most of the surgical tool tracking algorithms are based on machine vision methods,and the computational complexity is complex,and the real-time and accuracy cannot fully meet the actual needs.This article is based on practical application topics,in-depth study of endoscopic instrument tracking.Firstly,the endoscopic instrument detection algorithm is studied,including template matching method based on traditional feature extraction and YOLOv2 detection method based on deep learning.At the same time,in order to further improve the real-time and wide application of the algorithm,this thesis studies based on YOLOv2 and KCF.The endoscopic instrument tracking algorithm utilizes KCF to accelerate the tracking algorithm.For the detection of endoscopic instruments,this thesis studies the detection algorithm based on template matching and the detection algorithm based on YOLOv2.Firstly,the endoscopic instrument detection algorithm based on template matching was studied.In terms of algorithm accuracy,the SURF feature is combined with the gray histogram feature to improve the detection accuracy compared with the single feature.In terms of algorithm speed,this thesis proposes multi-step grayscale histogram matching,which reduces the detection time by changing the step value from large to small.In terms of algorithm speed,this thesis proposes multi-step grayscale histogram matching,which reduces the detection time by changing the step value from large to small.In order to further improve the detection accuracy and real-time performance,this thesis studies the endoscopic instrument detection algorithm based on YOLOv2,elaborates the training process and detection process of constructing the network,and builds the data set for training and testing.The template matching method improves the detection accuracy and real-time.In order to further improve the real-time performance and wide application of the algorithm,this thesis proposes an endoscopic instrument tracking algorithm based on YOLOv2 and KCF.Firstly,the KCF tracking algorithm is studied,and the KCF algorithm is improved accordingly.The CN color feature is added to better model the target.The multi-scale transform is added to better adapt to the target size change,and the tracking accuracy is improved.The GPU accelerates the KCF algorithm.Then,the improved KCF algorithm and YOLOv2 algorithm are combined to realize the target tracking algorithm of detection-tracking mode through grouping.This thesis studies the grouping parameter selection and the special case of tracking target disappearing.Finally,the test set is used to verify that the algorithm has good accuracy and real-time.Experiments show that the endoscopic instrument detection algorithm based on YOLOv2 has good detection accuracy and robustness,and it can also have good real-time performance for GPUs with better performance.The tracking algorithm based on YOLOV2 and KCF proposed in this thesis has better real-time and wide application,and has lower requirements on hardware,while ensuring the running speed without losing tracking accuracy.Finally,this thesis designs and implements the automatic navigation system of the robotic arm with mirror,expounds the design scheme from two aspects of hardware and software.Then summarize the work of the full text and determine the next research direction. |
