Karl C. Kluge

Extracting road curvature and orientation from image edge points without perceptual grouping into features

The ARCADE (Automated Road Curvature And Direction Estimation) algorithm estimates road curvature and tangential road orientation relative to the camera line-of-sight. The input to ARCADE consists of edge point locations and orientations extracted from an image, and it performs the estimation without the need for any prior perceptual grouping of the edge points into individual lane boundaries. It is able to achieve this through the use of global constraints on the individual lane boundary shapes derived from an explicit model of road structure in the world. The use of the least median squares robust estimation technique allows the algorithm to function correctly in cases where up to 50% of the input edge data points are contaminating noise. Two applications of ARCADE as the first stage of processing for a lane sensing task are described: 1) the extraction of the locations of the features defining the visible lane structure of the road; and 2) the generation of training instances for an ALVINN-like neural network road follower.

Performance evaluation of vision-based lane sensing: some preliminary tools, metrics, and results

The growth in the number of vision-based lane sensing algorithms in the literature has far outpaced the development of methods for characterizing the limits of performance of such algorithms. The large amounts of data such systems will need to correctly process to be sufficiently reliable for commercial deployment requires that tools be developed for the evaluation process which are almost completely automated. In order to gain some insight into the issues involved, a small pilot study was performed to compare the reliability of two methods used by the YARF road tracking system for locating white painted stripes in small image windows. The evaluation methodology and results are described. In addition, several proposed metrics for measuring road detectability are defined and evaluated.

Statistical characterization of the visual characteristics of painted lane markings

Most vision-based systems for lane detection and tracking use painted lane markings as the visual cues which determine the location of the camera relative to the lane. Almost all of the work that has been done in the area of evaluating the performance of these systems has focused on the accuracy of the recovered lane geometry. Reliability of feature detection as a function of intrinsic marking properties, ambient lighting and weather conditions, and viewing geometry is an equally important aspect of algorithm performance which must be explored if progress is to continue in this area of research. This paper reports a small scale effort to attack one aspect of this problem, the automated characterization of the intrinsic visual properties of white painted lane markings. Images of the right lane marking are taken by a camera mounted in an trailer enclosure towed behind a vehicle, allowing control of the lighting conditions. The intensity histogram of each image is examined to select a threshold which is used to classify each pixel as pavement or stripe. The edges of the white stripe are located using robust estimation and a shared vanishing point constraint. Once the stripe edges are located in an image, stripe properties such as width, brightness, and contrast with the pavement are calculated.

Lane Boundary Detection Using Deformable Templates: Effects of Image Subsampling on Detected Lane Edge

In order to be robust, a system for detecting road lane boundaries in images needs to be able to handle scenes which contain large amounts of clutter due to shadows, puddles, oil stains, skid marks, leaves and dirt on the road, etc. Many prior systems threshold the image gradient magnitude to detect edges, and then use the detected edge points to identify the lane boundaries. This can result in a very noisy edge image, as there are many situations in which there are strong edges in the image due to irrelevant clutter. There are also many situations in which the edges of the features defining portions of the lane boundaries have a lower intensity gradient than distracting clutter edges in the image. The LOIS (Likelihood of Image Shape) lane detection algorithm solves this problem by using a deformable template approach that uses image intensity gradient information in a way which does not require thresholding. This paper describes the LOIS algorithm in detail, and shows the results of applying the algorithm to a number of challenging scenes. Images are included comparing the lane boundaries detected by applying LOIS to a 240×256 image with the lane boundaries detected by applying the algorithm to a 30×32 subsampled version of the same image. Qualitatively the results are very similar, but the algorithm runs 45 times faster on the subsampled images.

Integrated Premission Planning and Execution for Unmanned Ground Vehicles

Fielding robots in complex applications can stress the human operators responsible for supervising them, particularly because the operators might understand the applications but not the details of the robots. Our answer to this problem has been to insert agent technology between the operator and the robotic platforms. In this paper, we motivate the challenges in defining, developing, and deploying the agent technology that provides the glue in the application of tasking unmanned ground vehicles in a military setting. We describe how a particular suite of architectural components serves equally well to support the interactions between the operator, planning agents, and robotic agents, and how our approach allows autonomy during planning and execution of a mission to be allocated among the human and artificial agents. Our implementation and demonstrations (in real robots and simulations) of our multi-agent system shows significant promise for the integration of unmanned vehicles in military applications.

TRACKING LANE AND PAVEMENT EDGES USING DEFORMABLE TEMPLATES

Experiments with the LOIS (Likelihood Of Image Shape) Lane detector have demonstrated that the use of a deformable template approach allows robust detection of lane boundaries in visual images. The same algorithm has been applied to detect pavement edges in millimeter wave radar images. In addition to ground vehicle applications involving lane sensing, the algorithm is applicable to airplane applications for tracking runways in either visual or radar data. Previous work on LOIS has focused on the problem of detecting lane edges in individual frames. This paper describes extensions to the LOIS algorithm which allow it to smoothly track lane edges through maneuvers such as lane changes.

Integrated premission planning and execution for unmanned ground vehicles

Fielding robots in complex applications can stress the human operators responsible for supervising them, particularly because the operators might understand the applications but not the details of the robots. Our answer to this problem has been to insert agent technology between the operator and the robotic platforms. In this paper, we motivate the challenges in defining, developing, and deploying the agent technology that provides the glue in the application of tasking unmanned ground vehicles in a military setting. We describe how a particular suite of architectural components serves equally well to support the interactions between the operator, planning agents, and robotic agents, and how our approach allows autonomy during planning and execution of a mission to be allocated among the human and artificial agents. Our implementation and demonstrations (in real robots and simulations) of our multi-agent system shows significant promise for the integration of unmanned vehicles in military applications.

Information assimilation research at the University of Michigan for the ARPA unmanned ground vehicle project

The goal of ARPAs Unmanned Ground Vehicle project is to demonstrate the use of small teams of cooperating autonomous robots (2 - 4 vehicles) to carry out military tasks in an outdoor environment. The role of the University of Michigan within the project focuses on aspects of mission planning, assimilation of information provided by multiple agents, and the interaction between planning and perception. The two aspects of this work related to sensor fusion are planning observation points to maximally reduce hypothesis uncertainty, and information sharing in multivehicle scenarios to reduce the amount of perception required. Observation point planning combines the systems current knowledge about an object with the uncertainty model used to characterize observations for data fusion in order to select optimal points for additional observations. Information sharing selects those detected features in the environment which are predicted to be most useful to other cooperating vehicles in the future, adding them to the multiagent systems model of the environment while ignoring less useful features.

STARLITE : A steering autonomous robot's lane investigation and tracking element

The problem of determining the offset to lane markings is an important one in designing vision-based automotive safety systems that operate on structured road environments. The lane offset information is critical for lateral control of the automobile. In this paper, we investigate the use of this information for an autonomous robots lane-keeping task. We employ a deformable template-based algorithm for determining the location of lane markings in visual images taken from a side-looking camera. The matching criteria involves a modification of the standard signal-to-noise (SNR) ratio-based matched filtering criteria. A KL-type color transformation is used for transforming the RGB channels of the given image onto a composite color channel, in order to eliminate some of the noise. The standard perspective transformation is used for transforming the offset information from image coordinates onto ground coordinates. The resulting algorithm, named STARLITE is robust to shadows, specular reflections, road cracks, etc. Experimental results are provided to illustrate the performance of STARLITE and compare its performance to the AURORA algorithm, and the SNR-based matched filter.