Hand Gesture Recognition System Using the Dynamic Vision Sensor

Abstract

With the rapid development of computer vision and artificial intelligence, human-computer interaction has become an inevitable part of people’s lives. Gestures can bring more natural, comfortable, and effective communication between people and machines. However, in some complex scenarios, such as rooms with looming lighting, the robustness and universality of hand gesture recognition based on traditional cameras are insufficient, and the supporting algorithms tend to underperform in real-time, especially for embedded devices. This article explores methods of implementing gesture recognition based on Dynamic Vision Sensor (DVS). We obtained frame data by event-based accumulation and time-based accumulation. Then we apply preprocessing techniques such as sub-time window and overlapping frame to achieve higher accuracy on hand gesture recognition with the DVS. In this paper, we built a DVS-based gesture recognition system with the advantages of efficient data preprocessing, low memory cost, low latency, and competitive recognition ability in interaction scenarios. The recognition accuracy reaches 94.6%.