Fleck, Tobias (FZI Research Center for Information Technology), Burger, Julian (FZI Research Center for Information Technology), Zöllner, J. Marius (FZI Research Center for Information Technology; KIT Karlsruhe In)

Adaptive Extrinsic Calibration of Roadside Cameras for Surveillance Tasks

Scheduled for presentation during the Regular Session "S03c-Connected Vehicle Technologies and Intelligent Infrastructure Systems" (WE-LA-T3), Wednesday, November 19, 2025, 16:40−17:00, Southport 3

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 19, 2025

Abstract

Precise perception and localization of traffic par- ticipants in roadside cameras is of major interest for multiple use cases: surveillance tasks, traffic motion analysis, smart traf- fic lights, and collaborative perception for connected automated driving. The precondition for precise localization of traffic ob- jects is good camera-to-world calibration, which can be altered by environmental influences over time, making maintenance and manual recalibration approaches costly and not scalable for offline surveillance tasks and completely impossible for online perception tasks. In this paper, we present two calibration schemes: the first calibration scheme uses aerial images to perform extrinsic calibration of roadside cameras towards a geo-referenced world coordinate frame. The second algorithm performs an online recalibration towards the world frame by estimating a homog- raphy from automatically detected correspondence points. We evaluate the presented algorithms using real-world data of an urban intersection recorded during a three day period including sun, dusk, night and rain. We give absolute and relative world errors metrics and discuss the error effects.