Minchae Lee

Local Path Planning for Off-Road Autonomous Driving With Avoidance of Static Obstacles

In this paper, a real-time path-planning algorithm that provides an optimal path for off-road autonomous driving with static obstacles avoidance is presented. The proposed planning algorithm computes a path based on a set of predefined waypoints. The predefined waypoints provide the base frame of a curvilinear coordinate system to generate path candidates for autonomous vehicle path planning. Each candidate is converted to a Cartesian coordinate system and evaluated using obstacle data. To select the optimal path, the priority of each path is determined by considering the path safety cost, path smoothness, and path consistency. The proposed path-planning algorithms were applied to the autonomous vehicle A1, which won the 2010 Autonomous Vehicle Competition organized by the Hyundai-Kia Automotive Group in Korea.

Enhanced Road Boundary and Obstacle Detection Using a Downward-Looking LIDAR Sensor

Detection of road boundaries and obstacles is essential for autonomous vehicle navigation. In this paper, we propose a road boundary and obstacle detection method using a downward-looking light detection and ranging sensor. This method extracts line segments from the raw data of the sensor in polar coordinates. After that, the line segments are classified into road and obstacle segments. To enhance the classification performance, the estimated roll and pitch angles of the sensor relative to the scanning road surface in the previous time step are then used. The classified road line segments are applied to track the road boundaries, roll, and pitch angles by using an integrated probabilistic data association filter. The proposed method was evaluated with the autonomous vehicle A1, which was the winner of the 2010 Autonomous Vehicle Competition in Korea organized by the Hyundai-Kia automotive group. The proposed method using the estimated roll and pitch angles can detect road boundaries and roadside, as well as road obstacles under various road conditions, including paved and unpaved roads and intersections.

Multiple exposure images based traffic light recognition

This paper proposes a multiple exposure images based traffic light recognition method. For traffic light recognition, color segmentation is widely used to detect traffic light signals; however, the color in an image is easily affected by various illuminations and leads to incorrect recognition results. In order to overcome the problem, we propose the multiple exposure technique which enhances the robustness of the color segmentation and recognition accuracy by integrating both low and normal exposure images. The technique solves the color saturation problem and reduces false positives since the low exposure image is exposed for a short time. Based on candidate regions selected from the low exposure image, the status of six three and four bulb traffic lights in a normal image are classified utilizing a support vector machine with a histogram of oriented gradients. Our algorithm was finally evaluated in various urban scenarios and the results show that the proposed method works robustly for outdoor environments.

Information Fusion of Cameras and Laser Radars for Perception Systems of Autonomous Vehicles

A autonomous vehicle requires improved and robust perception systems than conventional perception systems of intelligent vehicles. In particular, single sensor based perception systems have been widely studied by using cameras and laser radar sensors which are the most representative sensors for perception by providing object information such as distance information and object features. The distance information of the laser radar sensor is used for road environment perception of road structures, vehicles, and pedestrians. The image information of the camera is used for visual recognition such as lanes, crosswalks, and traffic signs. However, single sensor based perception systems suffer from false positives and true negatives which are caused by sensor limitations and road environments. Accordingly, information fusion systems are essentially required to ensure the robustness and stability of perception systems in harsh environments. This paper describes a perception system for autonomous vehicles, which performs information fusion to recognize road environments. Particularly, vision and laser radar sensors are fused together to detect lanes, crosswalks, and obstacles. The proposed perception system was validated on various roads and environmental conditions with an autonomous vehicle.

Overall Reviews of Autonomous Vehicle A1 - System Architecture and Algorithms

Abstract This paper describes an autonomous vehicle A1 that won the Hyundai Motor Group 2012 Autonomous Vehicle Competition. The A1 was developed for autonomous on-road and off-road driving conditions without driver intervention. The autonomous driving system consists of four parts that are localization, perception, planning, and control. Localization which estimates the ego-vehicle position on the map should be first performed to autonomously drive the A1. A perception algorithm detects and recognizes the objects around the ego-vehicle are also important to prevent collisions with obstacles and road departure. A planning algorithm generates the drivable motion of the A1 based upon previous information from the localization and perception system. Subsequently, a vehicle control algorithm calculates the desired steering, acceleration and braking control commands based on the information from the planning algorithm. This paper also presents the entire system architecture that the A1 used to accomplish all the required missions in the 2012 Autonomous Vehicle Competition.

Probabilistic lane detection and lane tracking for autonomous vehicles using a cascade particle filter

This paper proposes a robust lane detection algorithm with a cascade particle filter that incorporates a model decomposition approach. Despite the sophisticated tracking mechanism of a particle filter, the conventional particle-filter-based lane detection system suffers from an estimation accuracy problem and a high computational load. In order to improve the robustness and the computation time for lane detection systems, the proposed cascade particle filter decomposes a lane model into two submodels: a straight model and a curve model. By dividing the lane model, not only can the computation time be decreased, but also the accuracy of the lane state estimation system can be increased. The proposed lane detection algorithm and the cascade particle filter were evaluated on various roads and environmental conditions with the autonomous vehicle A1, which was the winner of the 2010 and 2012 Autonomous Vehicle Competition in the Republic of Korea organized by the Hyundai motor group. The proposed algorithm proved to be sufficiently robust and fast to be applied to autonomous vehicles as well as to intelligent vehicles for improving the vehicle safety.

A Task Scheduling Strategy in a Multi-core Processor for Visual Object Tracking Systems

Abstract : The camera based object detection systems should satisfy the recognition performance as well as real-time constraints. Particularly, in safety-critical systems such as Autonomous Emergency Braking (AEB), the real-time constraints significantly affects the system performance. Recently, multi-core processors and system-on-chip techno-logies are widely used to accelerate the object detection algorithm by distributing computational loads. However, due to the advanced hardware, the complexity of system architecture is increased even though additional hardwares improve the real-time performance. The increased complexity also cause difficulty in migration of existing algorithms and development of new algorithms. In this paper, to improve real-time performance and design complexity, a task scheduling strategy is proposed for visual object tracking systems. The real-time performance of the vision algorithm is increased by applying pipelining to task scheduling in a multi-core processor. Finally, the proposed task scheduling algorithm is applied to crosswalk detection and tracking system to prove the effectiveness of the proposed strategy.