Unusual Event Detection Using Machine Learning Techniques

Vivek Shenoy Y , Ganesh Hathwar B S , Sandesh Prabhu , Shubham M Vernekar

CNN, surveillance, algorithm, accuracy, deep learning, training, testing.

At the context, situation, due to variation of definitions between normality and abnormality unusual event detection for video sequences is a tough challenge. Though, it may generally be taken into consideration that unusual event or an activity, by way of unusual events happens much to a smaller extent compared to normal events. In order to get rid of unusual activities in surveillance videos, numerous kinds of modelling approaches are proposed inside the literature, inclusive of trajectory-based models, feature-based models and sparse reconstruction-based version, the main aim of these proposed models was to address the ambiguity in event detection and to examine and extract the hand-crafted-models or deep-learning-based models. Currently, the existing unusual-event detection models using deep learning techniques focuses on data represented using the feature form, which gives a slight importance to the impact of the internal structure characteristics of feature vector. Additionally, it is bit difficult to ensure the classification accuracy using single classifier. In order to overcome the above issues, we proposed an unusual event detection system using machine learning techniques using CNN algorithm. First, in order to extract the spatiotemporal features of video frame We use convolutional neural network (CNN) and long short-term memory models. This helps us to process the constructed model at the faster rate, and then in order to capture the internal sequential and topological relational characteristics of structured feature, the feature expectation for each key frame of every video is done. Finally, the constructed model is tested along with the trained model experiments by providing the video dataset as input. Comparison with some developed models the performance of this proposed method is better than several the state-of-the-art approaches.

Volume 7 Issue 7, July-2022

Pages : 1361 - 1366