Time-Spatial Feature Fusion for Contact Detection in Sports Using Multi-Sensor Data

The detection of athlete contact is of great significance for the safety of athletes and the fairness of sports events. The commonly used method is to have athletes wear sensors and set up multi-camera positions on the field, and then perform contact detection by analyzing the data from the sensors and cameras. This paper focuses on the scenario of dual video streams and multiple sensor data, where the images are sampled over time to obtain information features related to the temporal dimension. Multi-layer perceptron and XGBoost are used to fit the sensor data, and the EfficientNetB4 model is selected to extract features from multi-angle images stacked along the time dimension, thus achieving feature fusion based on both the time and spatial dimensions. The experimental results demonstrate the effectiveness of the proposed method in contact detection using multi-sensor data.