Action classification and analysis during sports training session using fuzzy model and video surveillance

Abstract

Activity recognition and classification are emerging fields of research that enable many human-centric applications in the sports domain. One of the most critical and challenged aspects of coaching is improving the performance of athletes. Hence, in this paper, the Adaptive Evolutionary Neuro-Fuzzy Inference System (AENFIS) has been proposed for sports person activity classification based on the biomedical signal, trial accelerator data and video surveillance. This paper obtains movement data and heart rate from the developed sensor module. This small sensor is patched onto the user’s chest to get physiological information. Based on the time and frequency domain features, this paper defines the fuzzy sets and assess the natural grouping of data via expectation-maximization of the probabilities. Sensor data feature selection and classification algorithms are applied, and a majority voting is utilized to choose the most representative features. The experimental results show that the proposed AENFIS model enhances accuracy ratio of 98.9%, prediction ratio of 98.5%, the precision ratio of 95.4, recall ratio of 96.7%, the performance ratio of 97.8%, an efficiency ratio of 98.1% and reduces the error rate of 10.2%, execution time 8.9% compared to other existing models.