An Adaptive Real-time Gesture Detection Method Using EMG and IMU Series for Robot Control

With the development of technology, people constantly put forward new requirements to the level and quality of human-computer interaction. In view of the intuitive and natural characteristics of sign language, gesture has become an important means of human-computer interaction. Therefore, a dynamic gesture recognition system which could adaptively recognize multiple gestures is the main development trend of gesture recognition system in the future. Though vision based dynamic gesture recognition system has always been one of the hotspots in the current scientific research field, its vulnerability to interference of light and obstacles is still a significate weakness that need to be overcome before utilizing in the real scenario. On this account, the current research constructs an actual time gesture recognition system fusing the physiological signal including Inertial Measurement Unit (IMU) and Electromyography (EMG), combined with adaptability design through self-adjusting threshold. Together, by evaluating experiment in 5 days, the adaptive classifier successfully identified 91.95% of 360 gestures with the latency of 10.89 ms, and the robots correctly responded to 88.61% of gestures in control dictionary.