Abstract
We present a trajectory mapping algorithm for a distributed camera setting that is based on statistical homography estimation accounting for the distortion introduced by camera lenses. Unlike traditional approaches based on the direct linear transformation (DLT) algorithm and singular value decomposition (SVD), the planar homography estimation is derived from renormalization. In addition to this, the algorithm explicitly introduces a correction parameter to account for the nonlinear radial lens distortion, thus improving the accuracy of the transformation. We demonstrate the proposed algorithm by generating mosaics of the observed scenes and by registering the spatial locations of moving objects (trajectories) from multiple cameras on the mosaics. Moreover, we objectively compare the transformed trajectories with those obtained by SVD and least mean square (LMS) methods on standard datasets and demonstrate the advantages of the renormalization and the lens distortion correction.
Highlights
Monitoring large areas such as airports and underground stations requires a set of distributed cameras to capture common patterns of activities and detect unusual events or anomalous behaviors [1]
We present a framework that embeds lens distortion correction into a homography estimation algorithm that attains the theoretical accuracy bound in homography estimation by minimizing residual misalignments
We demonstrate the proposed approach for trajectory transformation with lens distortion correction on the ETISEO
Summary
Monitoring large areas such as airports and underground stations requires a set of distributed cameras to capture common patterns of activities and detect unusual events or anomalous behaviors [1]. In most works involving trajectory mapping [2, 4,5,6,7], the authors adopt the linear pinhole camera model which assumes the principle of collinearity [8]. This model is only an approximation of the real camera projection. We present a framework that embeds lens distortion correction into a homography estimation algorithm that attains the theoretical accuracy bound in homography estimation by minimizing residual misalignments.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have