Deep-3DConvNet: A Network to Detect Abnormal Activities at Megastores

These days, there has been a rapid increase in cases of abnormal human behavior at megastores/shops, where people commit theft by stealing, consuming, or unwrapping packets when no one is seeing and then leaving the place without paying. Such unusual actions cause huge losses in business. Therefore, there is an urgent need to attract the research community's attention to detect abnormal events at megastores. To address this issue, we have designed an advanced three-dimensional convolutional neural architecture to identify abnormal activities at megastores. The proposed network is 15 layers deep, takes a video stream of resolution 120× 120 as input, and produces classification results as output. It extracts fine-tuned as well as general details from the video feed using small and large-sized 3D convolutional filters and categorizes them into respective classes. The proposed architecture is trained and tested on a synthesized action dataset that consists of human actions distributed into five classes: normal, stealing, eating, drinking, and damaging acts. Experimental results show that our model outperforms other state-of-the-art approaches with an accuracy of 88.88%.