top of page


Mr. John Smith

Job title



KEYWORDS – Autonomous Driving System, Contour, Convex Hull, L-shape, Confidence Score With the advancement of autonomous driving technology, the system is increasingly taking over control of driving maneuvers on behalf of the driver. To implement advanced autonomous driving system, sensor components such as camera, radar and LIDAR are essential for perceiving the surrounding environment. Among these sensors, LIDAR can provide accurate shape information of objects with high resolution in complex environments. However, the large number of points from LIDAR requires an efficient perception algorithm because of the limited computation resources in vehicles. Therefore, point selection algorithm that can represent object’s shape is necessary to ensure memory optimization and real-time processing speed. In the signal processing system for LIDAR point cloud, the contour point is used to express the shape of the object. The convex hull algorithm has been widely used in extracting the contour points. However, this algorithm has a limitation in that it cannot express various boundaries which have the concave shapes. This study presents a new algorithm for extracting contour points that can expresses various types of objects in highways and urban areas. In addition, this study proposes a confidence score that indicates the stability of L-shape or I-shape features when correcting the object’s direction of movement. The score can be used to evaluate the effectiveness of shape fitting. To extract the precise orientation angle of bounding box, five properties are used as constraints in calculating the confidence score. The presented method has been verified through experiments, with tests conducted in highway and Seoul urban environments. In order to represent shapes using a limited number of points, the length between points and the angle of adjacent line segments are used as constraints. It was confirmed that the proposed algorithm can represent the outlines of complex shapes such as concave shapes and inner edges. Furthermore, the experimental data shows that the confidence score provides a quantitative indicator for selecting the most robust contour shape among multiple contour layers. This is because the score reflects the features of the vehicle’s side surface that represent its forward movement. As a result, the proposed method demonstrates better performance in expressing the contours of objects with complex shape than conventional methods, making it possible to achieve low heading error. In this paper, a unified method was used for contour extraction, without distinguishing between dynamic and static objects. If class information about the object is available, proposed constraints of length and angle conditions can be applied with different parameters, allowing for more precise extraction of contour points that match the characteristics of each object. This means that by adjusting the parameters according to the object class, it is possible to achieve better contour extraction results. Moreover, the proposed method can be easily extended to handle various types of objects, making it applicable to a wide range of LIDAR perception applications. This study proposes a LIDAR-based object shape analysis technology suitable for autonomous driving environments, which can contribute to improving the performance of important applications such as collision avoidance and precise map matching in autonomous driving. In addition, the contour point extraction method can be extended to camera and radar sensors for extracting outer points of detected objects, which can have a significant effect on improving autonomous driving perception performance. In this paper, the contour point extraction technique was applied as a method for representing the outlines of objects detected by LIDAR. Through experiments extracting contour points of complex shapes such as guardrails on highways and buildings in urban areas, it was confirmed that the proposed algorithm is capable of representing not only the convex shapes of objects but their concave shapes. Additionally, an evaluation indicator was proposed to quantify the reliability of the contour shape when correcting the orientation of objects. By incorporating the confidence score, it was possible to reduce bounding box heading errors, resulting in improved performance.

Ms. Mirim Noh, Engineer, Hyundai motor company

Contour Extraction from LIDAR Point Clouds for Autonomous Vehicles

FWC2023-ITS-004 • Intelligent transport systems


Sign up or login to the ICC to download this item and access the entire FISITA library.

Upgrade your ICC subscription to access all Library items.

Congratulations! Your ICC subscription gives you complete access to the FISITA Library.


Retrieving info...

Available for purchase on the FISITA Store


bottom of page