Wearable Sensor-Based Human Behavior Understanding and Recognition in Daily Life for Smart Environments

Behavior recognition using motion sensors is getting prominence over other systems such as e-healthcare and life-log analysis systems especially in the healthcare domain for improving life expectancy and healthcare access. Accelerometers have been used in smart environments to recognize behavior since the last decade but heavy computation involved in recognizer model made them less acceptable. This paper proposed a computationally less expensive model with better recognition results for improved human behavior understanding system. Hierarchical features are used to ensure robustness as a performance attribute in the proposed system. These hierarchical features involve statistical features like signal magnitude, abrupt changes, and temporal variation among coordinates. Moreover, the extracted features are examined through the process of learning, training, and symbolization with the help of linear support vector machine. The examination of our recognition results based on feature extraction strategy show that our model excels others in terms of accuracy and computation time. The proposed system should be considered as a recommendation for systems involving human behavior recognition i.e. kindergarten, elderly at old-age houses and patients with Parkinson diseases.