Enhanced Abnormal Event Detection in Surveillance Videos Through Optimized Regression Algorithms
Jyothi Honnegowda, Komala Mallikarjunaiah, Mallikarjunaswamy Srikantaswamy · Journal of Intelligent Systems and Control · 2024
The recognition of abnormal events in surveillance video streams plays a crucial role in modern security systems, yet conventional techniques such as Support Vector Machines (SVMs) and decision trees (DTs) exhibit limitations in terms of accuracy and efficiency.These traditional models are often hindered by high false alarm rates and struggle to adapt to dynamic environments with variable conditions, thus reducing their practical applicability.In response to these challenges, an innovative approach, termed Adaptive Regression for Event Recognition (ARER), has been developed, leveraging advanced regression algorithms tailored for video data analysis.The ARER model integrates deep learning techniques, allowing for more precise temporal and contextual analysis of video footage.This methodology is structured through a multi-layered architecture that progresses from basic motion detection to complex anomaly identification.Trained on an extensive dataset covering a range of environmental and situational variables, ARER demonstrates enhanced robustness and adaptability.Evaluation results indicate that the ARER model achieves a 0.35% improvement in detection accuracy and a 0.40% reduction in false positives when compared to SVMs.Additionally, system throughput is increased by 0.25%, and detection latency is reduced by 0.30% in comparison to DTs.These advancements highlight the ARER approach as a superior alternative for real-time monitoring, offering significant improvements in both reliability and performance for surveillance applications.