Processing may take a few seconds...



Accurate Camera Pose Estimation from Learned Point Features: A Case Study


[ 1 ] Wydział Automatyki, Robotyki i Elektrotechniki, Politechnika Poznańska | [ SzD ] doctoral school student

Scientific discipline (Law 2.0)

[2.2] Automation, electronics and electrical engineering

Year of publication


Chapter type

chapter in monograph / paper

Publication language


  • camera pose estimation
  • keypoints
  • deep learning

EN This paper provides a case study of using a recent deep learning based approach to keypoint detection on images in the task of camera pose estimation. The application context is assisted docking to a charging station with an electric bus using monocular vision. We examined the influence of three factors on the achieved results: the backbone network, the size of the final activation maps generated by the network, and the number of convolutional layers in the keypoint head. The proposed configurations were evaluated to find the best trade-off between pose estimation accuracy (2D translation and the yaw angle estimation error were measured) and the computational complexity. The evaluation dataset was gathered using a real bus, during different weather conditions, and the ground truth data was provided by a Differential GPS. The result presented in this paper shows that proposed by us modifications of architecture can improve the accuracy of the whole processing system.

Pages (from - to)

98 - 102



Proceedings of the 3rd Polish Conference on Artificial Intelligence, April 25-27, 2022, Gdynia, Poland

Presented on

3rd Polish Conference on Artificial Intelligence, 25-27.04.2022, Gdynia, Polska

Points of MNiSW / chapter


This website uses cookies to remember the authenticated session of the user. For more information, read about Cookies and Privacy Policy.