Autonomous football exercise system based on convolutional neural network

Abstract

In order to build an intelligent autonomous football exercise system, this paper proposes a multi-level three-dimensional full convolutional network to segment the neuron football image according to the characteristics of neuron football image. In order to alleviate the problem that the segmentation prediction result generated by the network is biased to the background area and the foreground area is lost or only part of the foreground area is detected, the dice coefficient is introduced to calculate the overlap between the foreground class and the standard class and maximize this. For three[1]dimensional neuron football sports images, in order to detect three[1]dimensional breakpoints more conveniently, first analyze the two-dimensional slices. In addition, for two-dimensional neuron football motion image slices, two-dimensional high-curvature points are detected by using the covariance matrix eigenvalues of points on the curve segment and used as the initial screening of breakpoint candidate points. Finally, this paper applies the convolutional neural network to the autonomous football exercise system, researches the robot's action recognition and machine vision, builds an autonomous football exercise system based on the convolutional neural network, and analyzes and simulates its process. The research shows that the autonomous football exercise system based on convolutional neural network proposed in this paper has a certain degree of intelligence.