A real-time video surveillance system for traffic pre-events detection

Area of Science:

• Computer Vision
• Traffic Engineering
• Artificial Intelligence

Background:

• Road safety is a critical concern, with traffic violations contributing to accidents.
• Existing video surveillance systems often detect only one or two traffic pre-events.
• A comprehensive system for detecting multiple traffic anomalies is needed to improve road safety.

Purpose of the Study:

• To propose a conceptual framework for a video surveillance system to improve road safety.
• To develop and evaluate algorithms for detecting multiple traffic pre-events.
• To introduce the detection of 'wrong drop-off location of passengers' as a novel feature.

Main Methods:

• Development of a conceptual framework for video surveillance-based road safety.
• Implementation of five distinct algorithms for detecting traffic pre-events: speed violation, one-way traffic, overtaking, illegal parking, and wrong passenger drop-off.
• Validation of algorithms using 132 traffic videos from an Indian integrated plant and benchmark datasets (CamSeq01, ISLab-PVD).

Main Results:

• A single system capable of detecting five different traffic anomalies simultaneously.
• Demonstrated superiority of the developed algorithms compared to state-of-the-art methods in pre-event detection.
• Successful detection of 'wrong drop-off location of passengers'.

Conclusions:

• The proposed framework and algorithms significantly enhance road safety by enabling early detection of multiple traffic violations.
• The developed system offers a more comprehensive solution than previous systems, detecting five anomalies in one.
• The algorithms show strong performance and potential for real-world traffic management applications.