AI-Powered Surveillance
Debabrata Dansana, Abhijeet Joshi, Tanmaya Kumar Bhoi, Anil Kumar D, Vivek Kumar Prasa · 2024
Video surveillance systems detect suspicious activities and enhance security in public spaces. This paper employs deep learning models, specifically Long-term Recurrent Convolutional Networks (LRCN) and ConvLSTM, to recognize abnormal behaviors from video data. The LRCN model achieved 94% accuracy, 92.59% precision, 91.03% recall, and 91.55% F 1 score in detecting activities like running, fighting, and vandalism. The ConvLSTM model showed competitive results with 88.73% accuracy. Future research could expand to additional behaviors, integrate facial detection, and improve object tracking. This approach demonstrates deep learning’s potential to enhance public safety by automatically detecting abnormal activities.