Safe and efficient traffic signal operations rely on accurate detection of all road users, including vulnerable road users (VRU). However, conventional inductive loops struggle to detect pedestrians and push bikes as they have no or little amount of metal. To address this challenge, the ITS and Electrical Technology team evaluated computer vision technology in the field since early 2000. Computer vision is an emerging technology to derive meaningful information from digital images or videos, which has proven to be reliable and accurate. Thanks to the recent advancements in artificial intelligence (AI), computer vision technology gained new capability in learning and identifying the diverse profiles of VRUs including pedestrians, bikes, scooters, and tricycles with minimal processing power.
This paper elucidates the learnings and success of the use of computer vision in traffic detection. This paper covers the brief journey of how Transportation and Main Roads (TMR) adopted the technology and the local experience. It starts from the initial trial and subsequent deployment of computer vision detectors. It covers the utilisation of computer vision in smart pedestrian crossings for enhancing road throughput. The testing and evaluation of various computer vision detectors are also discussed. The side-by-side test results between computer vision and inductive loop detectors are presented to demonstrate the benefits, accuracy and reliability of this new technology. It also explores the feasibility and potential applications to utilise computer vision detectors in active transport and road safety.
