Machine Learning for Activity Recognition from Movement Time Series Data

Activity recognition is the core technology of intelligent video surveillance systems and focuses on the behavior of a single object or a pair of objects. A common approach to activity recognition is to first extract and track the characteristics of moving objects from an image sequence, with the goal of converting pixel-level data into low-level functions appropriate for activity analysis. And in these methods, it is important to successfully extract good features and analyze the spatio-temporal interaction of the type of object moving from the extracted features. In this paper, we propose a three step activity recognition model that takes into account the spatial elements of the trajectory using time series data of moving pedestrian trajectories. Activity recognition modeling used the c4.5 algorithm and performed recognition experiments on the four behaviors defined in the CAVIAR dataset.