The need for intelligent video analysis is in increasing demand as video surveillance is becoming more widespread. Manually looking through vast amount of video is tedious, time consuming, and human observers are known to be error prone, and subject to fatigue.
A sought-after feature of intelligent video analysis is to only extract sequences of interest, defined as rare cases which deviate from the norm; this is variously known as anomaly or abnormality detection. Such systems could raise alarms in case of anomalous activity, with the potential to allow for early and even preventive measures for various cases such as a violence, fires or ghost driver, etc. This is, however, so far an unsolved task in computer vision, due to complexity and context dependency.
The goal of this project is to advance the field of anomaly detection in order to make these methods feasible to apply in a wide range of applications.
This PhD project is directly funded as part of the Milestone Research Programme at AAU (MRPA).