Citation

Singh G, Khosla A, Kapoor R. J. Comput. Technol. Appl. 2019; 10(3): 19-28.

Copyright

(Copyright © 2019, STM Journals)

DOI

unavailable

PMID

unavailable

Abstract

It is significant to detect violent actions in video surveillance systems automatically, for example, bus stands, malls and railway stations. Though, the earlier detection techniques generally extract statistic features around the spatiotemporal interest points or extract descriptor in the regions where movement takes place, leading to limited abilities to successfully detect violence activities in video surveillance systems. To solve this problem, a new approach for the automatic detection of violence in video surveillance systems is proposed in this paper. Our proposed algorithm first extracts salient regions in the frames using PFT and Temporal PFT; salient regions may have a possible candidate region of violence. Then these salient region images are fed to the pre-trained deep networks to extract features and these extracted features are fed to the AdaBoost Support Vector Machine classifier for detection of whether the image contains any violent candidate or not. Experimental analysis on the two challenging standard datasets validates that the method proposed in this paper is better than the previous methods.

KEYWORDS: AdaBoost support vector machine, CNN, deep networks, PFT, Temporal-PFTCite this Article: Gajendra Singh, Arun Khosla, Rajiv Kapoor. Salient Region Guided Deep Network for Violence Detection in Surveillance Systems. Journal of Computer Technology & Applications. 2019; 10(3): 19-28p.

Language: en

Keywords

AdaBoost Support Vector Machine; CNN; Deep Networks; PFT; Temporal-PFT