Published Paper Details:

Detection of abnormal activities is essential but demanding research problem in computer vision. Anomaly detection in video surveillance involves monitoring of the designated area of interest to detect anomalies. Video surveillance is extensively used for a variety of fields, for instance, traffic monitoring, medical monitoring, security guarding, etc. Amid these various research fields, the detection of anomalous events plays a significant role. The goal is to identify abnormal behavior through the use of reliable features by utilizing deep learning methods. To address that, a fusion approach is proposed in videos based on three-dimensional convolutional neural networks (C3D) for extracting spatiotemporal features and convolutional long-short-term-memory (ConvLSTM) to aggregate the features that enhances the accuracy and efficiency of the model. The popular and open source UCSD dataset consisting of two sub datasets UCSD Ped1 and UCSD Ped2 has been used to evaluate the performance of the proposed method. The result demonstrates that we get high accuracy as compared to other approaches which is 94.3% and 92.6% for the first and second datasets respectively.