Research on the Application of Full Convolutional Neural Networks in the Accurate Detection of HD Video Targets in Sports Events

In order to overcome the shortcomings of low accuracy of object detection in existing machine vision, this project intends to use full convolutional neural network to detect the movement of athletes in sports video. The method of sampling, quantization and coding is adopted to realize the digitalization of motion video recording. By smoothing and enhancing sports video and extracting target sports video, the purpose of pre-processing sports video is finally achieved. This project intends to use a full convolutional neural network with jump connectivity and five concave convolutional neural networks with different expansion rates. A variety of jump link methods are used to extract multiple low-level features to achieve the significance edge extraction of multiple objects. Experimental results show that the accuracy and recall rate of moving image and still image are close to 1, indicating that this method has higher detection accuracy and less time.