Depth Estimation Using Particle filters for Image-based Visual Servoing

Abstract

In this paper, we present a novel approach for depth estimation in image-based visual servoing. Depth information are directly used in the control law to generate control signal, i.e. the screw velocity of the robot end-effector. Because rough estimates of depth values are not enough, we are motivated to this proposal. This approach employs a particle filter algorithm to estimate the depth of the image features online. A Gaussian probabilistic model is employed to model depth distribution. A set of depth particles is drawn in the current camera frame. The image measurements are used to recover the 3D samples. These samples are propagated to the next frame and projected into the image space. The maximum likelihood of 3D samples is the most probable to be the real-world 3D point. The mean value and the variance of the depth distribution are obtained from the maximum likelihood. The variance values converge to very small value within a few iterations. This gives high level of stability to the image-based visual servoing system. The simulation experiments show that the mean value goes very close to the real value of the depth in a few iterations. The depth is considered as the mean value of estimated distribution.