Abstract:

In public places, Abnormal activities are happening more frequently. The surveillance footages capture those actions. If an incident occurs, it will take a long time to watch the entire video in order to spot any unusual activity. So, by presenting a system to recognise activity within a fraction of time, one can save the time to detect an abnormal activity which is held at a specific moment. The proposed system keeps track of instances that show patterns of different human activities. This research focuses on a deep learning approach for detecting abnormal human activities in videos which include robbery, car accidents, and fighting. The system is implemented using pre-trained models such as VGG-16, ResNet50 and 2D-convolutional neural network (CNN). By merely providing a video as an input to the proposed system, it quickly recognises the abnormal activities present in the video

Description:

Identifying suspicious activity - robbery, road accident and fighting, Video surveillance system, VGG16, ResNet50, CNN