Human action recognition based on improved motion history image and deep convolutional neural networks
Qiuping Chun, Erhu Zhang · 2017
In order to make full use of video color image sequences for human action recognition, we proposed an approach to recognize human action based on motion history image (MHI) and deep convolution networks. Firstly, part of frames from the beginning and end of the motion video are removed and the gray MHI are extracted from the rest. Then we colorize them into 3 channels RGB by the rainbow encoding. Finally, we train a deep convolutional neural network to classify the RGB MHI by fine tuning a pretrained model. The experimental results indicate that the proposed method improves the recognition accuracy rate by 13% remarkably.