Enhancing CCTV Video Quality for Improved Segmentation and Detection of Suspicious Human Activities in Crowd and Non-Crowded Public Spaces: A Pre-Processing Approach

Abstract

Recently, reliance on surveillance systems to detect Suspicious Human Activities (SHA) in public and crowded places has increased. Despite the significant progress in surveillance cameras, climatic conditions and low light still challenge the quality of videos captured by these cameras. Moreover, the improvement methods vary with the diversity of challenges, such as low light and climatic conditions.

Therefore, researchers and developers resort to applying initial pre-processing for these videos to enhance the quality and contribute to their analysis. In addition, due to the lack of SHA datasets, in this work, we first collect a unique data set for SHA, which indicates the existence of a fight between a group of individuals in public places. Secondly, we propose a model to enhance the quality of videos using different techniques such as Adaptive Histogram, Gaussian Blur, Grayscale, and Median Filter. These techniques are considered the initial processing stage and prepare the dataset for detecting SHA in public and crowded places when the number of people is more than eight. In the future, we plan to apply segmentation techniques that enhance the performance of SHA detection.