Driven by vision. Defined by persistence.
B.S. in Automotive EngineeringKookmin University, Senior
Real-time environmental understanding through camera and LiDAR processing — detecting objects, segmenting lanes, and fusing multi-modal sensor data into a unified world representation.
Monocular camera feed processed through a real-time pipeline — lane detection via segmentation networks, object detection using YOLOv8 for vehicles, pedestrians, and traffic signs, and color-space-based preprocessing for varying lighting conditions.
Velodyne VLP-16 point clouds processed for ground plane segmentation, obstacle clustering via DBSCAN, and 3D bounding box estimation. Point cloud data is also used for occupancy grid generation feeding into the planning layer.
Camera and LiDAR data are fused in a common coordinate frame using extrinsic calibration. Projection of 3D LiDAR points onto the 2D image plane enables depth-augmented detection, improving both accuracy and reliability of environmental perception.