Research on motion recognition algorithm based on bag-of-words model

In this study, we proposed a motion recognition algorithm based on optimized temporal-spatial features and integrating BOW model. Firstly, detect the key points of the input video and then set up multiple 3D patches in the domain space, enhancing the spatiotemporal characteristic of the key points. With that, randomly sample the key point sets and carry out the description by histogram of oriented gradient (HOG) and optical flow histogram (OFH), thereby obtaining the composite descriptor for features. The clustering algorithm will be then applied to set up a visual dictionary and represent the input video samples. In the final step, a classifier is trained and generated by utilizing multi-core SVM; thus, the motion recognition has been completed. The experimental results suggested that the recognition rate and robustness of the proposed algorithm can outperform most existing motion recognition algorithms.