6533b862fe1ef96bd12c7359

RESEARCH PRODUCT

3D ACQUISITION SYSTEM APPLIED TO AGRONOMIC SCENES

subject

description

International audience; To improve results in automatic wheat ear counting by proxy-detection for early yield prediction, we need depth information of the scene. In this paper, we describe our 3D acquisition system dedicated to reconstruction of agronomic scenes. This system is composed of a camera mounted on a linear displacement driven by a microcontroller. The linear displacement allows acquiring a set of images in different distances to the scene. This image stack is used to apply shape from focus technique which is a passive and monocular 3D reconstruction method. This technique consists in the application of a focus measure for every pixel in the stack. An approximation method is used to refine the results of focus measure. Depth information of each point on the scene is determined by the maximum of focus measure. With depth information of each pixel, we create the depth map of the scene. The parameters previously inaccessible with 2D images like a leaf volume can be retrieve thanks to depth information.