<p>Real-time occlusion detection and prediction are critical to road safety and situational awareness in autonomous driving and advanced driver-assistance systems (ADAS). Occlusion is a condition when an object or region of interest is partially or fully occluded from the field of view of vehicle sensors, and this can lead to drastic blind spots in perception systems. This paper presents a full approach to occlusion detection and prediction based on a multisensor fusion technique with LiDAR, camera vision, and radar data. The approach integrates deep learning semantic segmentation and temporal tracking models to reason about the movement and position of occluded objects. A motion modeling and spatial-temporal correlation-based predictive algorithm is used to forecast potential occlusions before they happen. Experiments were conducted on large scales using the KITTI and nuScenes datasets, demonstrating a significant improvement in detection accuracy and prediction reliability. The results demonstrate the effectiveness of the approach in various traffic conditions, e.g., urban intersections, road intersections, and pedestrian zones. This work adds to making autonomous vehicle perception systems more safe, reliable, and intelligent.</p>