Early Detection of Collective or Individual Theft Attempts Using Long-term Recurrent Convolutional Networks
share
Theft crimes cause many losses to many facilities and companies around the world, this leads to a considerable number of risks, and despite the spread of a large number of surveillance camer-as, and large surveillance teams that track the movement that takes place within the store, the planning of many thieves can be Able to carry out the theft process without being noticed by human observers, as the human sight has movement limits, where the human observer can overlook one of the screens that records the actual movement at a moment, and therefore the theft can take place at this moment without Pay attention to it, and pre-planning the robbery can lead to the inability of human eyes to detect these attempts to carry out various theft operations. We proposed a model based on convolutional neural networks and recurrent neural networks to study the behavior and body language of shoppers within stores, where the proposed system can detect individual theft attempts or collective attempts to carry out the theft process. While other methods identify the crime itself, we instead model suspicious behavior - behavior that may occur before the accretion stage of crime - by exposing minute parts of the video with a high probability of containing the crime of shoplifting. Movement, which can lead to theft, the proposed neural structure was trained on a large number of visual clips that include attempts to steal according to a specific methodology. Through the proposed system, we reached an accuracy of 93 percent and a confidence coefficient of 93 percent.
READ FULL TEXT