Myeonghyeon Kim / Academic CV
From infrastructure perception to communication-assisted robot navigation: research prototypes, field experiments and system integration.
Integrated ROS 2, Nav2 and a hybrid safety node to make crossing decisions from signal phase and remaining green time. Compared SPaT and vision-based navigation under daytime, nighttime and dynamic occlusion conditions.
Investigating infrastructure-camera scene understanding for AMR waiting, entry and crossing decisions. The thesis plan covers pedestrians, vehicles, crosswalk occupancy and occlusion, with simulation and scenario-based evaluation.
Worked on LiDAR mapping, localization and Nav2 integration for mobile robots, including FAST-LIO navigation experiments at KAIST Munji Campus.
Configured an Isaac Sim delivery robot pipeline with LiDAR, IMU, camera, odometry, joint states and ROS 2 communication for navigation scenarios.
Prototyped an LLM context interpreter that adjusts rule-based PDM planner parameters. Implemented an uncooperative overtaking scenario in nuPlan and compared behavior with a mode-selection baseline.
Combined YOLOv8, DeepSORT, Re-ID and homography across five CCTV views to study consistent global IDs and pedestrian trajectory recovery.
Contributed to scenario design, GNSS trajectory collection and analysis of driving behavior around intersection blind spots.
Compared object detection models for real-time pedestrian recognition in infrastructure-based path prediction systems.
Developed motion detection with Gaussian Mixture Models, evaluated face detectors and integrated RetinaFace with WebSocket-based face recognition for automatic access control.
Developed vision-based driver gaze and hazardous-object detection components for warning systems. Work included gaze data acquisition and abnormal vehicle behavior analysis.