A Face Recognition Method for Sports Video Based on Feature Fusion and Residual Recurrent Neural Network

Abstract

Face recognition technology has penetrated into people's daily life and work fields, and has also been widely applied in sports videos. A video face recognition technology based on feature fusion and residual recurrent neural network is proposed to address the issue of image pose deviation caused by non-cooperative situations. Due to the large number of missing high-frequency data in low resolution facial images, a ternary adversarial reconstruction network is first proposed. It achieves correct image matching through the spatial distance of each image, improving the robustness of the model. For facial recognition in video sequences, higher precision key feature extraction is required. Therefore, this study introduced a residual recurrent neural network to optimize it, and designed its feature fusion and recognition network modules to compensate and extract relevant information before and after frames. Finally, performance verification analysis was conducted on the proposed model, indicating that the recognition accuracy of the recognition system reached 98.3%. In summary, the constructed residual recurrent neural network based on the ternary adversarial reconstruction network framework can effectively achieve video oriented facial recognition.