Design and implementation of traffic police hand gesture recognition system based on surface electromyographic signals

In the event of traffic congestion, unexpected traffic accidents, or severe weather, it is difficult to guarantee traffic safety and smoothness merely through traffic lights. Consequently, the traffic police are required to execute on-site command. However, it can be challenging for pedestrians to accurately notice and comprehend the hand gestures of traffic police in a complicated environment using only their eyes, which will result in incorrect judgments of the traffic situation. In order to minimize the influence of complex external environment on traffic police gesture recognition, we design and implement a traffic police hand gesture recognition system based on surface electromyography (sEMG) signals in this paper. In addition to establishing eight traffic police standard gesture datasets (TPSG) by the Arduino UNO development board and sEMG sensor, we also propose TSE-GRU, a novel neural network for accurate and robust traffic police gesture recognition. TSE-GRU incorporates the improved temporal convolutional network (TCN) and the gated recurrent unit (GRU). More specifically, the improved TCN employs the Squeeze-and-Excitation Networks (SE) that is modified to strengthen the representational power of temporal features from each TCN layers for extracting more advanced spatial features among multiple channel data, and the GRU captures long-term dependencies from time-series data. The experimental results show that TSE-GRU performs well and achieves 97.89% accuracy in the dataset TPSG under various experiment settings. The GUI interface of the recognition system can also show the current recognition results in real-time and timely provide feedback to the user on the traffic police gesture.