A Performance Analysis of Pre-trained Neural Network and Design of CNN for Sports Video Classification

Video content analysis and video classification are some of the very tricky problems among computer vision researchers. Especially in sports videos, shared common actions appear in different sports categories like jumping is a common action for both diving and high jump. These criteria need a tool or model to classify the video with an elevated rate of accuracy. Convolution neural network is an essential tool for deep learning. This is especially suited for image recognition and classification. Alexnet, googlenet and mobilenet are the examples for previously trained network. Feature extraction, classification, and transfer learning are the three primary functions of pre-trained networks. The main purpose of video classification in sports is to support sportspersons to find the video of their own interest in training and improve the performance. This paper analyzes the performance of the pre-trained network and the result is compared with a deep convolution network with customized layers on the same dataset. The experimental results are shown with graph and table format with the measure of accuracy and performance.