Monocular multi-target identification and positioning method based on artificial mark

Abstract

The invention discloses a monocular multi-target identification and positioning method based on artificial marks. The monocular multi-target identification and positioning method based on artificial marks includes the following steps: according to set colors, performing binarization processing on an original image so as to obtain a binary image; extracting contours after performing preprocessing on the binary image; performing ellipse fitting on each contour, and rejecting the ellipse which cannot satisfy the shape requirement; combining every two ellipses, and taking the endpoints of transverses as the apexes to form quadrangles; calculating the evaluation coefficient (i)e( / i) of each quadrangle; analyzing the evaluation coefficients, and performing multi-target selection; and taking four apexes of each quadrangle as the feature points, using an iterative method to solve PnP, and successively obtaining the three dimensional pose, of a camera, relative to each mark. The monocular multi-target identification and positioning method based on artificial marks uses identification of transverses to replace identification of segments, and provides an evaluation coefficient aiming at multi targets, thus being able to effectively screening marked targets and solving the relative poses even in a complicated background, and having high robustness. Besides, the monocular multi-target identification and positioning method based on artificial marks has the advantages of being simple and practicable, having relatively lower time complexity in the algorithm, being able to obtain a calculation result with high frame rate, and satisfying the requirement of high timeliness.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a monocular multi-target recognition and positioning method based on artificial markers. Background technique [0002] In the fields of industrial assembly, UAV fixed-point landing and augmented reality, it is necessary to use images for target recognition and pose estimation to guide the robot's actions. [0003] The difficulty of obtaining the target pose is mainly because the camera collects two-dimensional images, which will lose the depth information of the object. One solution is to use binocular vision, which can restore the depth information of the scene, but the calculation is complex and difficult to achieve on platforms with low computing power. Another solution is to use RGBD sensors to directly obtain depth information, but compared with monocular cameras, its cost is greatly increased. [0004] Target recognition and positioning based on artificial markers...

AI Technical Summary

Problems solved by technology

[0003] The difficulty in obtaining the target pose is mainly because the camera collects two-dimensional images, which will lose the depth information of the object

One solution is to use binocular vision, which can restore the depth information of the scene, but the calculation is complex and difficult to achieve on platforms with low computing power

Another solution is to use RGBD sensors to directly obtain depth information, but compared with monocular cameras, its cost is greatly increased

[0004] Target recognition and positioning based on artificial markers is a low-cost, easy-to-implement solution, but at present, most artificial markers require steps such as matching and encoding, and there are problems such as complex recognition process and low efficiency of multi-target detection

Images