Multi-Fusion Sensors for Action Recognition based on Discriminative Motion Cues and Random Forest

Through proper monitoring and recognition of actions of the human body many applications are possible in the areas of robotics, surveillance or personal health monitoring. Wearable inertial sensors and depth cameras significantly important for human action recognition results. Numerous computer vision methods are used for recognizing humanly actions but the performance is affected by location and orientation sensitivity or background clutter. In this paper, we proposed a novel approach to recognize the action, each action extract on the base of feature extraction method of time-domain, directional angle and depth motion map, for human action recognition (HAR) system. We extract different valuable features by different feature extraction methods. We used Random Forest Algorithm as a classifier with a benchmark UTD-MHAD dataset that achieves an accuracy of 90%. The results prove that the recognition provided by our proposed method is comparatively better in term of effectiveness and imprecision.