Recognition of complex human activities by using sequential pyramid matching

Human action recognition is a topic of increasing interest in recent years. Most of the work is focused on actions that have simple, periodic structure such as walking, running and hugging, but our everyday life contains very different types of actions with challenging problems. Space-time interest points and the bag of words approach have been shown a good performance on action recognition. For more complex activities, finer representations must be employed. In this article, we present the performance of the space-time interest points and bag of words approach by using sequential histogram pyramid on a challenging dataset. According to the test results, it is shown that in complex activity recognition the proposed approach increases recognition performance.