eduzhai > Applied Sciences > Engineering >

IoT System for Real-Time Near-Crash Detection for Automated Vehicle Testing

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: Our world is moving towards the goal of fully autonomous driving at a fastpace. While the latest automated vehicles (AVs) can handle most real-worldscenarios they encounter, a major bottleneck for turning fully autonomousdriving into reality is the lack of sufficient corner case data for trainingand testing AVs. Near-crash data, as a widely used surrogate data for trafficsafety research, can also serve the purpose of AV testing if properlycollected. To this end, this paper proposes an Internet-of-Things (IoT) systemfor real-time near-crash data collection. The system has several cool features.First, it is a low-cost and standalone system that is backward-compatible withany existing vehicles. People can fix the system to their dashboards fornear-crash data collection and collision warning without the approval or helpof vehicle manufacturers. Second, we propose a new near-crash detection methodthat models the target s size changes and relative motions with the boundingboxes generated by deep-learning-based object detection and tracking. Thisnear-crash detection method is fast, accurate, and reliable; particularly, itis insensitive to camera parameters, thereby having an excellenttransferability to different dashboard cameras. We have conducted comprehensiveexperiments with 100 videos locally processed at Jetson, as well as real-worldtests on cars and buses. Besides collecting corner cases, it can also serve asa white-box platform for testing innovative algorithms and evaluating other AVproducts. The system contributes to the real-world testing of AVs and has greatpotential to be brought into large-scale deployment.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...