Real world activity summary for senior home monitoring

It is a common knowledge that the daily activities of senior people tell a lot about their health condition. Thus, we believe that analysing their activities at home will improve the health care. Toward this goal, we propose a senior home activity summary system. There are two challenging problems in such a real world application. First, the amount of data for different activity categories is extremely unbalanced, which severely degrades the classifying performance. Second, senior's activities are usually accompanied by nurse's walking. It is impractical to predefine and label all the possible activities of all the potential visitors. Consequently, we propose a technique called subspace Naive-Bayesian Mutual Information Maximization (sNBMIM). It divides the feature space into a number of subspaces and allows the kernel and normalization parameters to vary between different subspaces. Moreover, we propose a novel feature filtering technique to reduce or eliminate the effects of the interest points that belong to other people. To evaluate the proposed activity summary system, we have collected a Senior home Activity Recognition dataset (UESTC-SAR), and performed activity recognition for eight categories. The experimental results show that the proposed system provides quite accurate activity summaries for a real world application scenario.