Researchers at NIT Rourkela develop an artificial intelligence-based multi-class vehicle detection (MCVD) model and light-fused bidirectional functional pyramid network (LFBFPN) tool to improve traffic management. Officials announced that. on friday.
The research team, led by Associate Professor Santos Kumar Das from the Department of Electronics and Communication Engineering, leveraged an intelligent vehicle detection (IVD) system that uses computer vision to identify vehicles in images and videos.
advertisement
They noted that the system collects real-time traffic data to optimize traffic flow, reduce congestion, and assist in future road planning.
The results of this research were published in the academic journal “Institute of Electrical and Electronics Engineers (IEEE) Transactions on Intelligent Transportation Systems.” The paper was published by Santos Kumar Das, researchers Prashant Deshmukh, Krishna Chaitanya Rayasam, Upendra Kumar Sahu of ECE, NIT Roukula, and Sudhan Majhi of Indian Institute of Science Bangalore. Co-authored by
The team developed a new MCVD model that uses video deinterlacing networks (VDnet) to efficiently extract key features from traffic images, even when vehicles vary in size and shape. We also introduced a specialized tool called Light Fusion Bi-Directional Feature Pyramid Network (LFBFPN) to further refine the extracted details.
“What is unique about LFFBPN is that it uses a simpler method and processes input data through a vehicle detection head (MVDH) without sacrificing accuracy, allowing accurate detection and detection of vehicles in all types of traffic situations. It’s about enabling classification,” Das said.
The MCVD model shows improved accuracy compared to existing methods. The real-time performance of the model was also evaluated on Nvidia Jetson TX2.
“By overcoming the limitations of older models and addressing the unique challenges of mixed traffic, the MCVD model provides a scalable option for real-time vehicle detection in developing countries,” Das concluded. .
