Citation Context

...t category, almost all approaches use tracking information directly to gather object speed and direction, and indirectly as an aid in determining features such as the size and aspect ratio of objects =-=[4, 14, 21, 24, 31]-=-. While trajectory based approaches are suitable for cases where the scene is comprised of only a few objects, in crowded environments it is difficult to reliably maintain tracks due to occlusion and ...

Citation Context

...Introduction The two major applications of camera surveillance are real-time crime prevention and crime investigation after an incident has occurred. For the first type of application event detection =-=[23, 22]-=- or aggression detection [6, 30] are used to understand people’s actions. For post incident investigation a complete reconstruction of the event is needed which additionally requires to follow persons...

Citation Context

...els (GMM) [10]; bayesian networks [11]; clustering [12]; fuzzy set theory [13]. Recently, the scientific community has been investigating anomaly detection for spatio-temporal data [14], trajectories =-=[15]-=- and moving object data [16], [17], [18]. Some of these approaches have been used to discover outliers for sea traffic data [19], [20], [21], [22], relying also on the structural properties of graphs ...

Citation Context

...r the temporal context that can be incorporated in the later processing. This improves robustness and enables analysis of subtle cues for instance in traffic model learning [11] and anomaly detection =-=[12]-=-. There are, however, challenges associated with trajectory grouping. First, trajectories often cover only a portion of the frames in the video sequence, due to occlusions, image noise, etc. As a resu...

Citation Context

...hniques could play an important role. In the computer vision area, behavior understanding studies can be classified into two categories: prominent activity recognition and abnormal behavior detection =-=[2]-=-. Prominent activity recognition is very difficult when the number of behavior models in an uncontrolled and uncooperative real-world data is considered [2]. On the other hand, abnormal behavior detec...

Citation Context

...nterest for traffic monitoring purposes [5], [6]. Persons are fundamental for human activity analysis [7], [8], then trajectories computed on such objects can be exploited for behaviour understanding =-=[9]-=-. Usually, video surveillance applications imply to pay attention to wide areas, hence different kinds of cameras are generally used, e.g, fixed cameras [10], omnidirectional cameras [11] or moving ca...

Citation Context

...ers created without the computation and storage demands of trajectory data mining techniques. Sequential pattern mining methods attempt to find common sequences of actions performed by moving objects =-=[17, 18, 19, 20, 21, 22, 23, 24]-=-. Finding commonly used driving routes is an example of sequential pattern mining. Sequential pattern mining is fundamentally different than clustering moving objects in that sequential pattern mining...

Citation Context

...d space, providing a favorable experimentation field for high-level video analysis methods [11]. Techniques relying on video trajectories have already been proposed for sports video activity analysis =-=[4, 21, 1, 10]-=-. Per˘se et al. proposed a basketball videos analysis method [19] that first segments tracked players trajectories into three different classes of basketball activities (offense, defense and timeout) ...

Citation Context

...l, by computing their degree of abnormality. Though simplistic in idea, this work was performed on a rather low resolution video in a relatively simple indoor scene of limited variability. Some works =-=[10, 8]-=- approach this problem by analysing object trajectories in video. Patterns can be extracted from trajectory data to understand the motion behaviour of objects in the video scene, and it is often neces...

Citation Context

...scuss only a subset of the past work here. Approaches tend to be divided between methods using tracked object paths and those that use features that do not rely on tracking. Systems that use tracking =-=[1,2,3]-=- attempt to extract the trajectories of objects of interest within a scene, then cluster those trajectories to identify outliers. Objects following96 C. De Leo and B.S. Manjunath unusual trajectories...