Detection of conspicuous behavior in street traffic by using B-splines as feature vector

Due to the increasing amount of data, a human operator might not be able to identify the important situations accurately. In order to improve the situation awareness of human operators in surveillance tasks, decision support systems need to direct the focus of the operators on situations of interests. These situations are often deviations from the typical patterns. Therefore, outliers and novelties have to be identified. In this paper, a datadriven algorithm for the detection of anomalies in trajectories based on b-splines is used to detect abnormal behavior in street traffic. The control points of a b-spline interpolation representing a trajectory are used as feature vector for anomaly detection algorithms. For the evaluation, two datasets of street traffic in cities are analyzed. In order to detect outlier in the datasets, the local outlier factor and the feature-bagging for outlier detection algorithm are used.