이곳에 짧은 설명을 써주세요

이곳에 짧은 설명을 써주세요

이곳에 짧은 설명을 써주세요

- Development of deep learning-based panoptic segmentation data from fish-eye camera
- Development and validation of the model using fish-eye camera dataset (WoodScape)
- Panoptic segmentation Focus on identifying classes for all pixels and distinguishing individual instances when they are foreground objects.
- Real time inference implementation using RTMaps

- Definition of data standards
- Define standards for AD dataset and metadata through analysis of large-scale domestic and international autonomous driving datasets
- Construction of semantic data extraction system
- Development of deep learning based semantic data automatic extraction system for AD dataset

- Lane-level route change detection based on lane map matching
- ML-based Lane-level Route Validation
- Precise map-based lane-level route validation using Machine Learning
- Route validation based on ML using surrounding perception information*
*Surrounding vehicle, road boundaries, lane detection, ego vehicle motion, etc.

- Calibration of various types of sensors
- Calibrate camera, LiDAR, and vehicle at once
- Pose graph construction between markers and sensors using an external camera

- Development of deep learning-based camera-LiDAR fusion model for point cloud panoptic segmentation (PCPS).
- Improvement of the performance of PCPS by continuously fusing intermediate features of camera and LiDAR
- Development and validation of the model using synthetic dataset (CARLA)
- Qualitative evaluation on the AEye 4Sight-M LiDAR point cloud

- Development of tracking algorithm using 4D imaging radar
- Development of lane-aided tracking algorithm using 4D imaging radar

- Development of RCS-LiDAR map generation using 4D imaging radar
- Development of RCS-weighted map matching localization algorithm using 4D imaging radar

- Creating a data augmentation module for adverse weather conditions.
- Analyzing drawbacks of current adverse weather augmentation methods
- Data Augmentation through statistical analysis of actual precipitation and wet ground noise
- Validation of augmentation module through actual adverse weather data
- Development of network for noise point and object classification
- Development of deep learning-based semantic segmentation network which robust to adverse weather
- Developing a multi-head precipitation classifier using point features

- Fisheye camera based online semantic segmentation
- Train various semantic segmentation network(RPNet, BiseNetv2) with Woodscape Dataset
- Deploy trained network using RTMaps

- LiDAR-video sensor synchronization
- synchronize 10Hz LiDAR signal with 30Hz EO-IR camera signal
- LiDAR-video sensor data calibration & sensor fusion
intrinsic/extrinsic calibration between LiDAR and EO/IR camera sensors
- depth extraction from 3D LiDAR data, fuse with 2D camera image
- Ground feature extraction and road surface analysis
- extract surface roughness, and road profile from foreground point cloud

- Viewpoint Change
- Difference between the mounting position of the camera sensor used in training dataset and the mounting position of the camera sensor used in inference process
- Causes performance degradation in image semantic segmentation
- Reduce the performance degradation of the model even if a viewpoint change occurs

- Dataset analysis
- Understanding and analyzing the latest trends in domestic and international autonomous driving-related - open datasets
Requirement establishment
Deriving specific dataset requirements according to development goals
Analysis of the need for constructing a large-scale domestic autonomous driving dataset through investigating the latest autonomous driving datasets

- Object classification based on multi-sensor fusion
- Developing an object classification algorithm using camera, LiDAR, and radar sensor data fusion
- Analyzing which sensor configuration and representation are effective for object classification
- Improving classification performance of misclassified objects
- Misclassification improvement through fusion of predictions from two distinct algorithms

- Improvement of point cloud semantic segmentation using low-level data from camera and LiDAR
- Improvement of classification of objects using semantic information and object information fusion