Camera Position Estimation with a Limited Number of Known Landmarks
Kim C.W.(1), Yi S.U.(2), Yoon W.S.(3), and Rhee S.A.(4*)

(1) Image Eng. Research Centre: Associate Research Engineer, 3DLabs Co. Ltd, Republic of Korea
(2) Image Eng. Research Centre: Assistant Research Engineer, 3DLabs Co. Ltd, Republic of Korea
(3) Image Eng. Research Centre: Research Engineer, 3DLabs Co. Ltd, Republic of Korea
(4*) Image Eng. Research Centre: Director of the Research Centre, 3DLabs Co. Ltd, Republic of Korea
*ahmkun[at]3dlabs.co.kr

Abstract

In crowded and complex indoor environments such as shopping malls and subway stations, individuals often find it difficult to navigate or determine their current location. The problem becomes more serious in environments surrounded by tall buildings or located deep underground, where Global Positioning System (GPS) signals are weak or completely unavailable. With the recent advancement of the geospatial information industry, interest in indoor spatial data development and indoor positioning technologies has been steadily increasing. This study proposes a method for estimating camera position using known points (landmarks) within images captured by a smartphone camera. In particular, we experimentally evaluate the feasibility of indoor positioning when only a limited number of known points is available. The proposed method begins by capturing indoor images with a smartphone camera. Simultaneously, external orientation parameters (roll, pitch, azimuth) are acquired through a custom Android-based application developed for this study. Ground truth data, including camera and landmark positions, are also collected. Next, we define a pinhole-based camera model for the smartphone. Given the limited number of known points, we assume that the camera and the landmarks lie on the same plane and accordingly fix certain parameters to simplify the model. Based on this camera model, observation equations are formulated and the parameters are estimated using the Least Squares Estimation (LSE) method. Although the proposed method does not reach the accuracy of approaches like SolvePnP, which use many landmarks to estimate both camera orientation and position, it still produces position estimates close to the ground truth in our experiments. The experimental results confirm that indoor position estimation is feasible with limited known landmarks. Future research will invesigate automation techniques such as AI-based object detection and automated landmark extraction, to further enhance the robustness and usability of the method.

Keywords: Indoor Positioning, Camera Pose Estimation, Landmark-based Localization

Topic: Topic B: Applications of Remote Sensing