Video Object Detection

Video object detection studies how to locate and classify objects across successive frames, where motion, occlusion, and changing viewpoints make single-frame detection unreliable. Common approaches extend image detectors with temporal feature aggregation, motion compensation, optical-flow or codec motion cues, and tracking-based propagation of detections. Benchmarks span settings such as small objects, aerial and satellite footage, RGB-thermal video, and crowded scenes, while edge-focused work uses adaptive frame skipping and compressed-domain inference to reduce cost. A central challenge is balancing temporal consistency and detection accuracy against latency and energy, especially under domain shift, camera motion, and difficult small-object conditions.