Concurrently detected and annotated abnormal events can contribute significant impacts on surveillance systems. While considering the specific domain of pedestrian trajectories, there are two main contributions of this paper. First, as introduced in a great amount of work of the trajectory-based anomaly detection literature, only the information of the pedestrian paths such as direction and speed is considered. Being differed from previous work, this paper proposes a framework that deals with an additional type of trajectory-based anomalies. This abnormal events happen when a person enters the prohibited regions. Those restricted regions are constructed by an online learning algorithm that uses surrounding information including detected pedestrians and background scenes. Second, a simple data-boosting technique is introduced as an aid for the lack of training data, such problem particularly challenges all previous work owing to the significantly low frequency of abnormal events. This technique only requires normal trajectories and fundamental knowledge of scenes to surge the number of training data for both normal and abnormal trajectories. With the increased number of training data, the conventional abnormal trajectory classifier is able to achieve better prediction accuracy without falling in the overfitting problem caused by complex learning models. Finally, the proposed framework, which annotates tracks entering prohibited areas, and a conventional abnormal trajectory detector using the data-boosting technique are integrated to form the united detector. Such detector faces with different types of anomalous trajectories in a hierarchical order. The experimental results show that all proposed detectors are able to effectively detect anomalous trajectories in the test phase.
Index Terms—Pedestrian detection, pedestrian tracking, anomalous trajectory, superpixel classification, trajectory features, Neural Network (NN).
Experimental result: video link