Static Gesture Recognition in the cabin Based on 3D-TOF and Low Computing Power

Traditional static gesture recognition algorithms are easily affected by the complex environment inside the cabin, resulting in low recognition rates. Compared with RGB photos captured by traditional cameras, the depth images captured by 3D-TOF cameras can not only reduce the influence of complex environments inside the cabin, but also protect crew privacy. Therefore, this paper proposes a low-computing static gesture recognition method based on 3D-TOF in the cabin. A low-parameter lightweight convolutional neural network (CNN) is used to train five gestures, and the trained gesture model is deployed on a low-computing embedded platform to detect passenger gestures in real-time while ensuring the recognition speed. The contributions of this paper mainly include: (1) Using the TOF camera to collect 1000 depth images of five gestures inside the car cabin. And these gesture depth maps are preprocessed and trained by lightweight convolutional neural network to obtain the gesture classification model. (2) In the gesture preprocessing stage, a method based on depth information is designed to quickly locate the depth range of the hand area, which can quickly locate the depth range of the hand area in real-time. (3) A low-parameter lightweight convolutional neural network model is proposed, which has fewer training parameters and can be deployed on a low-computing embedded platform. The experimental results show that compared with traditional static gesture recognition algorithms inside the cabin, this method has higher accuracy and stronger robustness and can recognize passenger gestures in real-time on a low-computing embedded platform.