This research work is focused on detecting visual anomalies/abnormal
events in video surveillance systems. In such situations, human attention may wander,
and abnormal situations may be more challenging to note. So, to avoid such situations,
an automated system is needed that can analyze such a huge amount of data and trigger
alarms in abnormal events. Nowadays, the automation of surveillance systems is a
major research area. This chapter presents a 3D local descriptor-based anomaly
detection method for traffic surveillance videos, capable of extracting information
about the appearance and motion of objects. Appearance information is extracted by
3D Histogram of Gradients (HOG), and motion information is extracted by 3D
Histogram of Optical Flow Orientation (HOOF). Finally, these features are fed to a
combined classifier for the detection of abnormality. Appearance information plays an
important role when different types of objects are in the scene, like in traffic
surveillance videos. This algorithm is tested on the YouTube video due to the
unavailability of a publicly available standard traffic surveillance dataset.
Keywords: Abnormality detection, Histogram of gradients, Optical flow orientation, Surveillance systems, Traffic surveillance.
