Optimization Method Of Circular Target Recognition And Tracking In Motion Capture

With the continuous development of computer vision,object detection has become one of the research hotspots in the field of computer vision.As an important subset of visual targets,circular targets are often found in everyday life and various industrial applications.In the motion capture system,more optical spherical markers(Marker points)are used to mark on moving objects,and the spatial position of optical markers is determined by the camera to achieve target recognition and motion capture,so as to obtain the motion trajectory of the object.Therefore,the identification of circular marker points has become very important,and has become a hot research direction in the field of motion capture.In the early detection part,the performance of anti-environmental interference,recognition accuracy,speed and stability still needs to be improved.The recognition result will directly affect the accuracy of the system to complete the subsequent motion capture work,so how to accurately and stably identify the Marker point in the more complex background has become an urgent problem to be solved.This thesis mainly focuses on the recognition and tracking of circular targets in motion capture systems,mainly including the following work content:Firstly,aiming at the problem of low stability of Marker point recognition of spherical targets in complex background when applied to motion capture system,this thesis proposes a semi-supervised training Marker point recognition method based on YOLOv5 s recognition detection technology: firstly,Teacher student model is used to make pseudo-labels,and highconfidence pseudo-labels are selected according to the obtained threshold.Then,the pseudolabel data is used as the training set to fine-tune the model to obtain the final model.Next,the Residual module in the CSP module in the YOLOv5 s model is replaced,the pruning operation of the model is carried out,and the module in the YOLOv5 s algorithm is deleted by using the dynamic capture small target recognition feature,so as to further realize the model lightweight.Finally,spherical Marker points are used to test in different scenarios,and the experimental results verify the effectiveness of the proposed method,which confirms that the proposed method can extract Marker points in complex scenes more stably,thereby enhancing the adaptability of the motion capture system in complex environments.Secondly,this thesis compares and discusses the traditional target tracking algorithm and the multi-target tracking algorithm around the multi-target tracking strategy,and finally selects the DeepSORT algorithm based on deep learning as the tracking strategy.Combine improved detection algorithms with DeepSORT algorithms for better performance.The data set captured by the non-contact six-degree-of-freedom motion measurement system is used for test evaluation,which proves that the algorithm can ensure the stability of the identity information of the detected object and complete the tracking work.Finally,aiming at the problem that optical motion capture system is not applicable in complex scenes,an optimization strategy based on deep learning is proposed,and its feasibility is verified.In this thesis,the images taken by a camera in both indoor and outdoor environments are used as the test set to test the optimization strategy.The experimental results show that the combination of the optimization recognition detection algorithm and the DeepSORT algorithm proposed in this thesis can meet the target recognition and tracking work of the optical motion capture system in the early stage,and then confirm the feasibility of the proposed optimization strategy.