Incorporating duration and region association information in trajectory classification
Dhaval J. Patel · Journal of Location Based Services · 2013
Trajectory classification is the process of predicting the class label of moving objects based on their trajectories and other features. Existing works on building trajectory classification model discover features by using spatial distribution and shape of sub-trajectory. However, they do not utilise duration and region association information available in trajectory data during feature generation. In this study, trajectory features are generated using spatial distribution, duration and region association information of trajectories. In particular, two types of features, region rules and path rules, are generated from trajectories for classification. Region rules consider the spatial distribution of trajectories, the time spent (duration) by the trajectories in the region and the association information with other regions. Path rules differentiate objects based on their travelling patterns and speed. Efficient algorithms are devised to obtain region rules and path rules. Based on the discovered rule, trajectory classification model is built to predict the class label of new trajectory. Experimental results on various real-world data-sets show that incorporating duration and region association information in trajectory classification improves accuracy.