Calibration of an embedded camera for driver-assistant systems

Abstract

Abstract-One of the main technical goals in the actual automotive industry is to increase vehicle safety. The European project SPARC (Secure Propulsion using Advanced Redundant Control) is developing the next generation of trucks towards this aim. The SPARC consortium intends to do so by providing the truck with active security systems. Specifically, by equipping the vehicle with different sensors, it can be made aware of its environment, such as other vehicles, pedestrians, etc. By combining all sensor data and processing it with internal proprioceptive information 1 , the truck can advice, warn or even override the driver in case of non-response.Camera systems are particularly advantageous for sensing purposes, because they are passive sensors and provide very rich information. Moreover, they can easily be software-reconfigured to extract new or additional data from the input-image. Typical information that SPARC aims to extract is the position of the vehicle within the lane, the presence and distance of other vehicles or obstacles, and the identification of roadsigns. In this paper, a lane-detection algorithm will be presented and discussed.Some of the resulting information needs to be given in world coordinates, as opposed to image coordinates. To carry out the necessary conversion, a previous calibration is needed. The challenge is to determine a procedure to calibrate a camera mounted on a truck to precisely determine the position of obstacles situated in a 100 meter range. The two-step calibration procedure presented here has been designed to simplify the calibration of the mounted cameras in the truck production line.

Efficient and robust extrinsic camera calibration procedure for Lane Departure Warning

“…Due to difficulties to guarantee a flat road or the technical feasibility for the production line, [8] proposes a simple geometry in front of the camera. Preferably, the pattern should cover the whole image.…”

A novel approach for the online initial calibration of extrinsic parameters for a car-mounted camera

“…After detecting the markers in the acquired images, a backprojection model is fitted by finding the describing parameters such as the three rotation angles and translational offsets. The main differences in offline calibration methods can be found in the choice of the fiducial markers, measurement technique, orientation of the calibration plane and the optimization algorithm [2], [3], [4], [5]. Additionally, many closed-form solutions have been proposed by several simplifications and linearization of the camera model [6], [7], [8].…”

Calibration and Sensitivity Analysis of a Stereo Vision-Based Driver Assistance System

“…As to lane-related applications, Bellino et al have used a single planar pattern and the aforementioned method for calibrating a monocular system used in heavy vehicles (Bellino et al, 2005). The authors used a fixed principal point and did not involve γ .…”

“…A rough pose estimation can be performed by first estimating the homography between the world plane and the image and then by re-using the orthogonality constraints with known camera calibration matrix K (Malm & Heyden, 2003;Lepetit & Fua, 2005). The planar pattern used in intrinsic calibration is unsuitable for far-range systems, because it minimizes errors for close-range (as also stated in Marita et al, 2006;Bellino et al, 2005). For this purpose, Broggi et al used a medium-range grid painted on the ground (Broggi et al, 2005) while Marita et al used vertical X-shaped markers placed on the ground in front of the vehicle in a distance up to 45 m (Marita et al, 2006).…”