Detection and Recognition of Suspicious Multitask Human Action Identification from Preloaded Videos using CCTV Stationary Cameras
Pavankumar Naik, Srinivasa Rao Kunte R · Journal of Machine and Computing · 2025
Even more emphasis has been made on the use of video surveillance for sighting suspicious activities in the common places. As with other retrospective investigations, forensic investigations and riot inspections have normally required the use of automated offline video processing systems. However, development in the area that attempts at real time event detection has not been very impressive. Thus, the present work aims at developing a framework for processing raw video data gathered by a stationary colour camera within a given area to allow for real-time analysis of the observed activities. The suggested strategy begins with the acquisition of Object-level data by following and identifying objects and people in the scene via blob matching in real-time. Temporal features of those blobs are used to semantically characterize behaviours and events in terms of object and interobject motion attributes. A number of behaviours that are pertinent to public safety, such as lounging, gatherings, fainting, fighting, stealing, abandoned objects, occlusion, Abuse, Arrest and other activities available on UCF crime dataset. Were selected for the purpose of this demonstration of this method. The conclusions suggested in the work are based on experiments performed with currently easily accessible libraries.