Threat Detection with Facial Expression and Suspicious Weapon

Abstract

Recently criminal activities such as robbery with the threat of life using weapons have increased exponentially. In the past, CCTV cameras were used for providing proof of criminal activities. But due to the technological advancements in areas like deep learning, image processing, etc., various ways are coming into the picture to prevent those criminal activities. Weapon detection was the first criteria introduced to define an activity as “suspicious”, but it had many drawbacks, such as dummy weapons were being classified as a threat and also the system failed where there was frequent use of weapons. As the phrase “Suspicious Activity” is a relative entity, human being can identify a friendly environment, where people are carrying weapons but the motto is not harmful. On the other hand, existing technology lacks this context. Here, we have tried to align the existing system of weapon detection and facial expression with this context. This enhances the ability of the system to take decisions as if it is thinking as a human brain. Our CNN model inspired by the VGGNet family named as suspExpCNN model optioned accuracy 66% was yielded for facial expression detection and Faster RCNN and SSD models yielded for weapon detection with accuracy 82.48% and 82.84%, respectively. The work further combined suspExpCNN model with weapon detection model and generated alert if an angry, scared or sad expression is detected.