Academic(Thesis)
Mono-scale CNN-LSTM fusion network for Suspicious Activity Detection Technology
Problem Statement
Developed a deep learning model to detect suspicious activities (robbery, shoplifting, fighting) in video footage by analyzing extracted frames, addressing the challenges of real-time monitoring in surveillance.
Key Contributions
Design and Development: Created a Mono-scale CNN-LSTM fusion model to process video frames, leveraging pre-trained CNNs (VGG16, ResNet) for spatial and temporal feature extraction.
Frame Extraction and Optimization: Extracted frames from videos, applied data augmentation, and used transfer learning to enhance detection accuracy.
Evaluation: Achieved high accuracy in detecting crime across diverse scenes, boosting the effectiveness of automated video surveillance.
Research: Documented methodologies and findings in a research paper, contributing to advancements in automated crime detection technology.