Robust Hand Detection Based on Convolutional Neural Network and Attention Module

Abstract

The hands are essential parts, helping people to contact and communicate with the surrounding environment. Hand gesture and position detection is an interesting topic in computer vision field, it was applied in the areas such as action recognition, Human-Computer Interaction, Human-Robot Interaction, control systems, etc. With the strong emergence of artificial neural networks and computer hardware devices, it becomes easier to apply hand detection in practice. Based on the benefits of convolutional neural network (CNN) and bottleneck attention module, this paper proposes a robust CNN for hand detection. The proposed network achieved 95.52% of average precision (AP) on the Egohands test set and 59.07 frames per second (FPS) on the Intel Core I7–4770 @ 3.40 GHz CPU in real-time testing.