Novel qualitative visual odometry for a ground: Vehicle based on funnel lane concept

Abstract

Visual odometry is the process of estimating the position of a robot from visual information. Visual odometry methods are classified as: appearance-based methods which use the intensity information of the image and feature-based methods which use the image features information. Feature-based methods are more common in visual odometry. Visual odometry methods usually need calibration, essential matrix calculations and finding minimum argument solution. In this paper, we present qualitative visual odometry estimation for a ground vehicle. The method requires a camera with no calibration and there is no need for essential matrix or either any minimum argument calculations. The proposed method is based on funnel lane concept that was presented to control a robot to move from a current image to a destination image. A funnel lane is created based on features geometric information and the robot is controlled in such a way that does not go outside the funnel lane until reaching the destination image. The idea in this paper is to do a reverse thing and instead of creating a funnel lane to control the robot to reach an image destination, we suppose that the robot is moving straight until it is satisfying funnel lane constraints and a turning angle is calculated when the constraints are unsatisfied. We show that this approach is giving significant results in indoor and outdoor environments and it can be a simple way to visual odometry.