University Archives

As part of the Electrical and Computer Engineering Senior Design course, this project develops the perception subsystem for a Level-5 autonomous ground vehicle competing in the International Ground Vehicle Competition (IGVC) Auto-Navigation Challenge. The perception subsystem detects lane boundaries, identifies obstacles, and classifies terrain to provide the vehicle with real-time environmental awareness. The design process began with a needs and metrics matrix derived from customer and competition requirements, which were translated into engineering objectives including detection accuracy, environmental adaptability, and real-time processing performance. The system integrates multiple sensing modalities, including camera-based vision and Light Detection and Ranging (LiDAR) sensors. Visual perception is performed using an Intel RealSense D456 depth camera, which provides synchronized red–green–blue (RGB) and depth information. Lane boundaries are detected using the You Only Look Once Panoptic version 2 (YOLOPv2) architecture, where only the lane-line segmentation head is used to extract lane boundary information from camera images. The depth and camera calibration information from the RealSense sensor are converted into three-dimensional spatial coordinates (x, y, z), enabling the perception system to estimate the positions of points in the environment relative to the vehicle. Obstacle detection is performed using You Only Look Once version 11 (YOLOv11) applied to the RGB images, and the detected obstacles are associated with the corresponding spatial coordinates obtained from the camera. These obstacle positions are then compared and confirmed using LiDAR measurements, improving the reliability of object localization and environmental perception. All perception modules are implemented within the Robot Operating System 2 (ROS 2) framework, where sensor streams, detection outputs, and spatial information are processed and shared across system nodes. The resulting perception pipeline enables the vehicle to continuously detect lane boundaries, recognize obstacles, and interpret the surrounding environment in real time to support autonomous navigation within the IGVC competition environment.