Abstract:
From Past few years, the vast numbers of cameras are installed in various public and
private areas for security, monitoring abnormal human activities, and traffic. The
detection and recognition of abnormal activity in a real-world environment is a big
challenge as there can be many types of alarming and abnormal activities like theft,
violence and accidents. This research deals with the accidental events in traffic videos.
As population is increasing drastically the likely hood of accidents is also increasing. In
modern world the video-based camera surveillance system (VCSS) is used for traffic
surveillance, monitor which is called video traffic surveillance cameras (VTSS). The
VTSS is used to detected abnormal events or incidents regarding traffic on different roads
and highways like traffic block, traffic congestion, and vehicle accidents. This research
proposes a methodology for detecting accidental events automatically through
surveillance videos.
Review of literature suggests that convolutional neural network (CNN) which is a
specialized Deep learning approach pioneered to work with Grid like data, is effective in
image and video analysis. This research uses CNN to find anomaly (accident) from videos
captured by VTSS. In training of CNN model, vehicle accidental image dataset (VAID)
composed of images with anomalies, is constructed and used. For testing the proposed
methodology, the trained CNN model is checked on multiple videos and results are
collected and analysed. The result of this research shows the successful detection of
traffic accident events at the rate of 80% in the traffic surveillance system videos.
