Detection of Violence in Videos using Hybrid Machine Learning Techniques

  • Sumitra Kisan, Baishnabi Sahu, Abhijit Jena, Sachi Nandan Mohanty


Surveillance videos are able to capture a variety of real-world anomalies. All kinds of activities like walking, talking, riding a vehicle are considered normal behaviour. But any sort of activity that does not adhere by the definition of normal pattern is considered as abnormal or anomalous. It can be any irregular behaviour like yelling, theft in public, breaking into a house, accident and many more. In this work, the intention is to develop a model that will take videos captured from CCTV cameras on streets, ATMs, police station, roads, hospitals, railway stations, etc., as input and classify them as normal or abnormal (contains anomaly at some point of time). A video which contains anomaly is labelled as positive and normal videos are labelled negative. Change in motion vector, is taken as the key for classification. This research, aims to find several spatial and temporal feature descriptors like Histogram of Oriented Gradients, Histogram of Optical Flow and then use the extracted features in classifiers such as CNNs for HOG and HOF. The aim the proposed method is to return positive hits for portions of video sequences that contain violent content.

How to Cite
Abhijit Jena, Sachi Nandan Mohanty, S. K. B. S. (2020). Detection of Violence in Videos using Hybrid Machine Learning Techniques. International Journal of Advanced Science and Technology, 29(3), 5386- 5392. Retrieved from