FACPM: Crop Hand Area with Attention

Hand tracking is an important part of Human-Computer Interaction (HCI). Computer vision based hand tracking can free the users and help HCI more natural. Hand area should be cropped from the image before hand tracking. Inspired by the success of deep learning, we use CNN to crop the hand area. We proposed a new method named FACPM to crop hand area with attention mechanism and Feature pyramid networks. We adopt VGG as base network, to improve the speed of the framework, unimportant network is cut off. FPN model is used for different scale features and can perform better on tiny feature. Attention model is used for distributing the weights of different features in FPN. A better cropped image can lead to a more precise 3D hand pose estimation and can be used in hand tracking.