Human Suspicious Activity Recognition

Abstract

Surveillance security cameras are progressively being used for monitoring purposes in almost every area, especially watching people and their behaviour for safety purposes. In criminology, photos taken from such devices are typically used to determine who may be the individuals involved when an accident happens. Although this cameras use is essential for a post-crime response, real-time surveillance is required to serve as an early alert to deter or stop an incident before it arises. We developed and implemented an early warning system in this project that automatically identify people in a surveillance camera environment where no human movement happens in a specified time period and then utilized data from our database to identify whether or not the person is authenticated. If not, we will alert the approved person or people by text message and give their Android application a snap of the specific location. Using convolutional neural networks we train a feature extraction model for face recognition to get a strong recognition performance on the Chokepoint data set obtained utilizing surveillance cameras. The approach also gives the purpose of sending suspicious location images to an android app for that we are creating a mobile application using the Android language.