Human Activity Recognition Based on the Combined SVM&HMM

Being able to recognize human activities is essential for several applications, including personal assistive robotics and smart homes. In this paper, we perform the recognition of the human activity based on the combined SVM&HMM in daily living environments. Firstly, we use a RGBD sensor (Microsoft Kinect) as the input sensor, and extract a set of the fusion features, including motion, body structure features and joint polar coordinates features. Secondly, we propose the combined SVM&HMM that combines the SVM characteristics that can reflect the difference between the samples with the HMM characteristics that is suited to deal with the continuous activities. The SVM&HMM plays their respective advantages of SVM and HMM comprehensively. Thus, the combined model overcomes the drawbacks of accuracy, robustness and computational efficiency comparing with the separate the SVM model or the traditional HMM model in the human activity recognition. We test our algorithm on recognizing twelve different activities performed by four people in different environments, such as a kitchen, a living room, an office, etc., and achieves good performance even when the person was not seen before in the training set. The experiment results show that our algorithm possess the better robustness and distinction.