Self-Calibration of Traffic Surveillance Cameras Based on Moving Vehicle Appearance and 3-D Vehicle Modeling

Na Wang, Haiqing Du, Yong Liu, Zheng Tang and Jenq-Neng Hwang

ICIP 2018

Summary

This method calibrates traffic-surveillance cameras using vehicle appearance and 3D vehicle models. Edge and line detection provide initial vanishing-point candidates, while clustering and regression reduce noise. Projected line segments from vehicle models refine the geometric estimates, and evolutionary optimization relaxes assumptions about camera parameters and vehicle trajectories. Experiments evaluate the approach across several traffic-video datasets.

Citation

@inproceedings{Wang18SelfCalVeh,
  author = {Na Wang and Haiqing Du and Yong Liu and Zheng Tang and Jenq-Neng Hwang},
  title = {Self-calibration of traffic surveillance cameras based on moving vehicle appearance and 3-D vehicle modeling},
  booktitle = {Proc. ICIP},
  pages = {3064--3068},
  address = {Athens, Greece},
  year = {2018}
}