The localization of autonomous vehicles using light detection and ranging (LiDAR) sensors relies on high-definition (HD) maps, which are essential for accurate positioning. However, the large storage capacity required for HD maps poses challenges for real-time performance. To address this issue, we propose a vehicle to infrastructure (V2I)-based LiDAR localization method. In this approach, real-time HD maps are transmitted to vehicles in the vicinity of the infrastructure, enabling localization without the need for map data. We conducted tests to determine the optimal size of the HD maps and the distance between vehicles and the infrastructure, considering the impact on transmission speed. Additionally, we compared the matching performance between the complete HD map and sub maps received from the infrastructure, to evaluate the effectiveness of our method in a qualitative manner.
In this study, we propose an enhanced LiDAR-based mapping and localization system that utilizes a camera-based YOLO (You Only Look Once) algorithm to detect and remove dynamic objects, such as vehicles, from the mapping process. GPS, while commonly used for localization, often fails in urban environments due to signal blockages. To address this limitation, our system integrates YOLOv4 with LiDAR, enabling the removal of dynamic objects to improve map accuracy and localization in high-traffic areas. Existing methods using LiDAR segmentation for map matching often suffer from missed detections and false positives, degrading performance. Our approach leverages YOLOv4’s robust object detection capabilities to eliminate potentially dynamic objects while retaining static environmental features, such as buildings, to enhance map accuracy and reliability. The proposed system was validated using a mid-size SUV equipped with LiDAR and camera sensors. The experimental results demonstrate significant improvements in map-matching and localization performance, particularly in urban environments. The system achieved RMSE (Root Mean Square Error) reductions compared to conventional methods, with RMSE values decreasing from 0.9870 to 0.9724 in open areas and from 1.3874 to 1.1217 in urban areas. These findings highlight the ability of the Vision + LiDAR + NDT method to enhance localization performance in both simple and complex environments. By addressing the challenges of dynamic obstacles, the proposed system effectively improves the accuracy and robustness of autonomous navigation in high-traffic settings without relying on GPS.
30 November 2018 in Journal of Institute of Control, Robotics and Systems
This paper proposes a design for a Leader-Follower tracked vehicle and its system architecture. The System consists of 3 parts: Perception, Navigation, and Control. Perception system is configured using a 1-channel LiDAR(light detection and ranging) and camera. By the use of these, Tracked vehicle recognizes the color of clothes who needs to be followed in various unstructured agricultural environments. Navigation system has been developed based on the concept of the Leader-Follower and the method of geometric path tracking (as studied in the robotics field). The vehicle can follow the Leader and be controlled by convergence of skid steering and pure pursuit (instead of using a complicated vehicle dynamics or land modeling process). The ultimate goal of this paper is explain the convergence process through formula and β compensation coefficient for steer is proposed for the formula. The β coefficient can compensate for the rotational ability of the vehicle without compensate modeling process. The influence of the β coefficient is investigated with actual tracked vehicle experiments.
1 June 2020 in Transaction of the Korean Society of Automotive Engineers
This paper proposes a method for creating a path for parking autonomous vehicles, regardless of the distance and the position of the parking space, by using a vehicle with sensor performance superior to the existing ultrasonic and camera-based sensors. Although conventional methods have a limited parking detection area, the proposed method has a greater free parking detection area. In addition, a new tracking algorithm was employed in order to follow the generated path. Meanwhile, a new algorithm was developed for the longitudinal controller to perform perfect parking in order to eliminate the ‘path following error’ caused by the difference between the steering speed and the vehicle speed. The path planning and path tracking method for the Auto Valet Parking System was then experimentally demonstrated.
1 June 2020 in Transaction of the Korean Society of Automotive Engineers
Unmanned vehicles require environmental awareness when moving from origin to destination, as well as a high semantic knowledge of the environment. This study deals with the removal of noise from point cloud maps used in the environmental recognition systems of unmanned vehicles. However, a moving object is unnecessary on a map. In this study, we define noise as unnecessary and we deal with noise removal systems. This study proposes to define the region of interest based on the driving line of an unmanned vehicle, object separation using local thresholds, improved clustering using Euclidean clustering and DBSCAN, and classification using the support vector machine(SVM). This method allows for the classification of solid lines, dotted lines, median strips, guardrails, and noise.
In the context of LiDAR sensor-based autonomous vehicles, segmentation networks play a crucial role in accurately identifying and classifying objects. However, discrepancies between the types of LiDAR sensors used for training the network and those deployed in real-world driving environments can lead to performance degradation due to differences in the input tensor attributes, such as x, y, and z coordinates, and intensity. To address this issue, we propose novel intensity rendering and data interpolation techniques. Our study evaluates the effectiveness of these methods by applying them to object tracking in real-world scenarios. The proposed solutions aim to harmonize the differences between sensor data, thereby enhancing the performance and reliability of deep learning networks for autonomous vehicle perception systems. Additionally, our algorithms prevent performance degradation, even when different types of sensors are used for the training data and real-world applications. This approach allows for the use of publicly available open datasets without the need to spend extensive time on dataset construction and annotation using the actual sensors deployed, thus significantly saving time and resources. When applying the proposed methods, we observed an approximate 20% improvement in mIoU performance compared to scenarios without these enhancements.