Quality and energy aware data acquisition for activity and locomotion recognition

With the advent of wireless sensors and pervasive environments, autonomic human activity recognition has received substantial attention in research. In such environments, many sensors are deployed on each object with the purpose to collect sufficient data to recognize the activities of the object. To perform activity recognition, low-level data streams from the sensors are combined at the sink. A key challenge is to recognize efficiently and with high accuracy the object's activities based on the low-level sensor data. However, there is a trade-off between high accuracy and efficiency, caused by the cost of delivering data samples from sensors to the sink. The challenge is to determine sampling rates that satisfy the required accuracy and minimizes the communication cost. We formalize this problem of choosing sampling rates that satisfy the required accuracy and minimize the communication cost. We formalize this problem as an integer programming problem and solve it by using Lagrangian relaxation with branch-and-bound method. Evaluation results with a publicly available dataset demonstrate the potential applicability of our approach.