ITERATIVE REFINEMENT FOR CAMERA POSE ESTIMATION BASED ON VISUAL MARKERS

Alkhatib Majdi Nasib, Bobkov Alexander Valentinovich

The use of computer vision is a very important problem for many practical applications related to the development and operation of autonomous mobile platforms, for example, logistics robots in automated production and warehousing. In this paper we implement an iterative visual navigation algorithm based on Gauss-Newton method for estimating position based on the image captured by only one camera. We find the pose that minimize the squared error of the predicted location of the image features to the measured location. For computing the error of pose estimation, synthetic images were generated. Considered a two-dimensional object (square) containing a set of distinct points (at least four points). The algorithm error is calculated for many relative distances between the camera and the object and three noise levels. Synthetic images of the body are generated and taken from different elevation and azimuth angles, the position of the camera in relation to the body is estimated by applying the algorithm, then compared with the real position, to calculate the error.

Keywords:iterative algorithm, synthetic images, image localization, visual navigation.