Human activity recognition based on multiple order temporal information

• Ubiquitous sensors are used to collect human activity data.
• Human activity recognition is modeled as a classification problem.
• Extract features combining pattern mining with discriminative ability descriptor.
• Features with low or high frequency have limited discriminative power.

Automatically recognizing human activity in daily life is of great importance to our society. However, this task is very challenging due to various factors, such as the inconsistent movement speed and duration from different people. In this paper, we model the problem of human activity recognition as a classification problem. Our model improves on previous methods through the definition of a representation scheme that uses multiple order temporal information. We also show that the features with low and high support have limited discriminative power. Based on this conclusion, our scheme consists of three steps: The first step is to select features by frequent pattern mining. The second step is to select features based on a novel discriminative power descriptor which is called sequence frequency-inverse activity frequency. The third step is to design classifier. Experimental results show that our solution scores good performance.

Figure optionsDownload as PowerPoint slide