Context analysis : sky, water and motion

Interpreting the events present in the video is a complex task, and the same gesture or motion can be understood in several ways depending on the context of the event and/or the scene. Therefore the context of the scene can contribute to the semantic understanding of the video. In this paper, we present our research on context analysis on video sequences. By context analysis we mean not only determining the general conditions such as daytime or nighttime, indoor or outdoor environments, but also region labeling [1] and motion analysis of the scene. This paper reports on our research results on sky and water labeling and on motion analysis for determining the context. Later, this can be extended with regions such as roads, greenery, buildings, etc. Experiments based on the above detection techniques show that we achieve results comparable with other state-of-the-art techniques for sky and water detection, although in our case the color information is poor. To evaluate results, we use the Coverability Rate (CR) which measures how much of the true sky or water is detected by the algorithm. The obtained average of CR for water detection is about 96:6% and for sky detection it is about 98%.