Video Risk Detection and Localization using Bidirectional LSTM Autoencoder and Faster R-CNN
Idir Boulfrifi, Mohamed Lahraichi, Khalid Housni · Advances in Science Technology and Engineering Systems Journal · 2021
This work proposes a new unsupervised learning approach to detect and locate the risks "abnormal event" in video scenes using Faster R-CNN and Bidirectional LSTM autoencoder.The approach proposed in this work is carried out in two steps: In the first step, we used a bidirectional LSTM autoencoder to detect the frames containing risks.In the second step, for each frame containing risks, we first used Faster R-CNN to extract all the objects containing in the scene and then for each object detected we check whether it represents a risk or not.In other words, in testing phase, the frames with events deviated from normal features learned in training phase are detected as risk.To locate objects representing risk, only the objects detected by Fast R-CNN deviated from normal feature are classified as risk.Experimental results demonstrate that the proposed method can reliably detect and locate the object representing risk in video sequences.