Perception System for Autonomous Vehicles: Object Detection, Motion Tracking, and Future Position Prediction

This paper presents a multi-sensor fusion approach for detecting, clustering, and predicting the motion of dynamic objects by integrating camera and LiDAR data. The framework combines YOLO-based visual segmentation with LiDAR point clouds to enable accurate spatial localization and motion tracking. Ego-motion compensation is achieved using IMU and velocity data, while trajectory prediction is enhanced through centroidbased orientation estimation. Detected objects are segmented into coherent clusters using DBSCAN, facilitating visualization and motion forecasting. The framework was validated on a simulated KITTI dataset, demonstrating its effectiveness in dynamic object detection and trajectory prediction.