Feature Based Training for Crime Detection using Deep Learning Techniques
P. Mahalakshmi, S. Thenmalar · 2023
Today’s surveillance systems produce a massive amount of video data every day, making it challenging for computer vision professionals to analyze it. Since they happen sporadically and with a low likelihood in actual surveillance, manually searching for odd events in these enormous video feeds is a difficult task. The public is better protected by anomaly detection using deep learning, which has more reliable detection capabilities and minimizes human labor. The framework for effective features-based intelligent anomaly detection presented in this research can function in video surveillance networks with a low time complexity. In the given framework, we take a set of frames in sequence and extract spatiotemporal properties from them by sending each frame to a pretrained Convolutional Neural Network (CNN) model that has already been trained. The collected features from the series of frames are useful for identifying unusual events. Then the Bi- directional Long Short-Term Memory (Bi- LSTM) model is used to classify anomalous and normal occurrences in various surveillance scenes once we pass in the feature maps that were retrieved.