• IEMEK Journal of Embedded Systems and Applications
• /
• v.16 no.3
• /
• pp.107-111
• /
• 2021

A 3D point cloud map is an essential elements in various fields, including precise autonomous navigation system. However, generating a 3D point cloud map using a single sensor has limitations due to the price of expensive sensor. In order to solve this problem, we propose a precise 3D mapping system using low-cost sensor fusion. Generating a point cloud map requires the process of estimating the current position and attitude, and describing the surrounding environment. In this paper, we utilized a commercial visual-inertial odometry sensor to estimate the current position and attitude states. Based on the state value, the 2D LiDAR measurement values describe the surrounding environment to create a point cloud map. To analyze the performance of the proposed algorithm, we compared the performance of the proposed algorithm and the 3D LiDAR-based SLAM (simultaneous localization and mapping) algorithm. As a result, it was confirmed that a precise 3D point cloud map can be generated with the low-cost sensor fusion system proposed in this paper.

• The Journal of Korea Robotics Society
• /
• v.18 no.3
• /
• pp.346-351
• /
• 2023

With the release of numerous open driving datasets, the demand for domain adaptation in perception tasks has increased, particularly when transferring knowledge from rich datasets to novel domains. However, it is difficult to solve the change 1) in the sensor domain caused by heterogeneous LiDAR sensors and 2) in the environmental domain caused by different environmental factors. We overcome domain differences in the semi-supervised setting with 3-stage model parameter training. First, we pre-train the model with the source dataset with object scaling based on statistics of the object size. Then we fine-tine the partially frozen model weights with copy-and-paste augmentation. The 3D points in the box labels are copied from one scene and pasted to the other scenes. Finally, we use the knowledge distillation method to update the student network with a moving average from the teacher network along with a self-training method with pseudo labels. Test-Time Augmentation with varying z values is employed to predict the final results. Our method achieved 3rd place in ECCV 2022 workshop on the 3D Perception for Autonomous Driving challenge.

• Journal of the Korean Society of Industry Convergence
• /
• v.26 no.6_3
• /
• pp.1325-1332
• /
• 2023

Recently, much research is being conducted based on point cloud data for the growth of innovations such as construction automation in the transportation field and virtual national space. This data is often measured through remote control in terrain that is difficult for humans to access using devices such as UAVs and UGVs. Drones, one of the UAVs, are mainly used to acquire point cloud data, but photogrammetry using a vision camera, which takes a lot of time to create a point cloud map, is difficult to apply in construction sites where the terrain changes periodically and surveying is difficult. In this paper, we developed a point cloud mapping system by adopting non-repetitive scanning LiDAR and attempted to confirm improvements through field application. For accuracy analysis, a point cloud map was created through a 2 minute 40 second flight and about 30 seconds of software post-processing on a terrain measuring 144.5 × 138.8 m. As a result of comparing the actual measured distance for structures with an average of 4 m, an average error of 4.3 cm was recorded, confirming that the performance was within the error range applicable to the field.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.17 no.4
• /
• pp.633-640
• /
• 2022

As service robots are applied to various fields, interest in an image processing processor that can perform an image processing algorithm quickly and accurately suitable for each task is increasing. This paper introduces an image processing processor design method applicable to robots. The proposed processor consists of an AGX board, FPGA board, LiDAR-Vision board, and Backplane board. It enables the operation of CPU, GPU, and FPGA. The proposed method is verified through simulation experiments.

• Journal of Auto-vehicle Safety Association
• /
• v.13 no.4
• /
• pp.14-19
• /
• 2021

This paper presents LiDAR static obstacle map based vehicle dynamic state estimation algorithm for urban autonomous driving. In an autonomous driving, state estimation of host vehicle is important for accurate prediction of ego motion and perceived object. Therefore, in a situation in which noise exists in the control input of the vehicle, state estimation using sensor such as LiDAR and vision is required. However, it is difficult to obtain a measurement for the vehicle state because the recognition sensor of autonomous vehicle perceives including a dynamic object. The proposed algorithm consists of two parts. First, a Bayesian rule-based static obstacle map is constructed using continuous LiDAR point cloud input. Second, vehicle odometry during the time interval is calculated by matching the static obstacle map using Normal Distribution Transformation (NDT) method. And the velocity and yaw rate of vehicle are estimated based on the Extended Kalman Filter (EKF) using vehicle odometry as measurement. The proposed algorithm is implemented in the Linux Robot Operating System (ROS) environment, and is verified with data obtained from actual driving on urban roads. The test results show a more robust and accurate dynamic state estimation result when there is a bias in the chassis IMU sensor.

• ETRI Journal
• /
• v.43 no.4
• /
• pp.603-616
• /
• 2021

Over the last few years, autonomous vehicles have progressed very rapidly. The odometry technique that estimates displacement from consecutive sensor inputs is an essential technique for autonomous driving. In this article, we propose a fast, robust, and accurate odometry technique. The proposed technique is light detection and ranging (LiDAR)-based direct odometry, which uses a spherical range image (SRI) that projects a three-dimensional point cloud onto a two-dimensional spherical image plane. Direct odometry is developed in a vision-based method, and a fast execution speed can be expected. However, applying LiDAR data is difficult because of the sparsity. To solve this problem, we propose an SRI generation method and mathematical analysis, two key point sampling methods using SRI to increase precision and robustness, and a fast optimization method. The proposed technique was tested with the KITTI dataset and real environments. Evaluation results yielded a translation error of 0.69%, a rotation error of 0.0031°/m in the KITTI training dataset, and an execution time of 17 ms. The results demonstrated high precision comparable with state-of-the-art and remarkably higher speed than conventional techniques.

• Journal of Cadastre & Land InformatiX
• /
• v.49 no.1
• /
• pp.17-27
• /
• 2019

In this study, we conducted a study to predict the photovoltaic power by constructing the sensor based meteorological data observation system and the accurate terrain data obtained by using LiDAR surveying. The average sunshine hours in 2018 is 4.53 hours and the photovoltaic power is 2,305 MWh. In order to analyze the effect of photovoltaic power on the installation angle of solar modules, we installed module installation angle at $10^{\circ}$ intervals. As a result, the generation time was 4.24 hours at the module arrangement angle of $30^{\circ}$, and the daily power generation and the monthly power generation were the highest, 3.37 MWh and 102.47 MWh, respectively. Therefore, when the module arrangement angle is set to $30^{\circ}$, the generation efficiency is increased by about 4.8% compared with the module angle of $50^{\circ}$. As a result of analyzing the influence of the seasonal photovoltaic power by the installation angle of the solar module, it was found that the photovoltaic power was high in the range of $40^{\circ}{\sim}50^{\circ}$, where the module angle was large from November to February when the weather was cold. From March to October, it was found that the photovoltaic power amount is $10^{\circ}{\sim}30^{\circ}$ with small module angle.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.33 no.6
• /
• pp.1099-1110
• /
• 2023

The recognition system of autonomous driving and robot navigation performs vision work such as object recognition, tracking, and lane detection after multi-sensor fusion to improve performance. Currently, research on a deep learning model based on the fusion of a camera and a lidar sensor is being actively conducted. However, deep learning models are vulnerable to adversarial attacks through modulation of input data. Attacks on the existing multi-sensor-based autonomous driving recognition system are focused on inducing obstacle detection by lowering the confidence score of the object recognition model.However, there is a limitation that an attack is possible only in the target model. In the case of attacks on the sensor fusion stage, errors in vision work after fusion can be cascaded, and this risk needs to be considered. In addition, an attack on LIDAR's point cloud data, which is difficult to judge visually, makes it difficult to determine whether it is an attack. In this study, image scaling-based camera-lidar We propose an attack method that reduces the accuracy of LCCNet, a fusion model (camera-LiDAR calibration model). The proposed method is to perform a scaling attack on the point of the input lidar. As a result of conducting an attack performance experiment by size with a scaling algorithm, an average of more than 77% of fusion errors were caused.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.21 no.5
• /
• pp.253-273
• /
• 2022

Automated vehicles rely on information collected through sensors to drive. Therefore, the uncertainty of the information collected from a sensor is an important to address. To this end, research is conducted in the field of road and traffic to solve the uncertainty of these sensors through infrastructure or facilities. Therefore, this study developed a traffic cone that can maintaing the gaze guidance function in the construction site by securing sufficient LiDAR detection performance even in rainy conditions and verified its improvement effect through demonstration. Two types of cones were manufactured, a cross-type and a flat-type, to increase the reflective performance compared to an existing cone. The demonstration confirms that the flat-type traffic cone has better detection performance than an existing cone, even in 50 mm/h rainfall, which affects a driver's field of vision. In addition, it was confirmed that the detection level on a clear day was maintained at the 20 mm/h rain for both cones. In the future, improvement measures should be developed so that the traffic cones, that can improve the safety of automated driving, can be applied.

• Journal of Drive and Control
• /
• v.19 no.4
• /
• pp.62-69
• /
• 2022

Construction machinery is exposed to accidents such as collisions, narrowness, and overturns during operation. In particular, mobile crane is operated only with the driver's vision and limited information of the assistant worker. Thus, there is a high risk of an accident. Recently, some collision avoidance device using sensors such as cameras and LiDAR have been applied. However, they are still insufficient to prevent collisions in the omnidirectional 3D space. In this study, a rotating LiDAR device was developed and applied to a 250-ton crane to obtain a full-space point cloud. An algorithm that could provide distance information and safety status to the driver was developed. Also, deep-learning segmentation algorithm was used to classify human-worker. The developed device could recognize obstacles within 100m of a 360-degree range. In the experiment, a safety distance was calculated with an error of 10.3cm at 30m to give the operator an accurate distance and collision alarm.