A Real-Time Multi-Object Tracker Based on Hybrid Integration of ByteTrack and DeepSORT Using YOLOv8

Real-time multi-object tracking (MOT) remains a challenging task due to the inherent trade-off between computational efficiency and robust identity preservation. Spatial-only methods such as ByteTrack offer high throughput by re-using low-confidence detections and simple IoU-based association, but they often suffer from identity switches when targets overlap or change appearance. Conversely, appearance-aware approaches like DeepSORT employ deep reidentification embeddings to maintain consistent identities under occlusion, at the expense of increased computational cost. In this paper, we propose a novel hybrid framework that integrates ByteTrack's efficient IoU-based matching with DeepSORT's appearance-driven association, underpinned by the YOLOv8 detector for high-quality, real-time object localization. Our unified matching cascade addresses parameter mismatches and resolves potential ID collisions by jointly leveraging spatial and visual cues. We validate the proposed system on the MOT17 benchmark, demonstrating a significant improvement over the individual baselines: the hybrid tracker achieves a MOTA of 0.6694 and an IDF1 of 0.6329-compared to 0.4531/0.3593 (MOTA/IDF1) for ByteTrack and 0.1187/0.4776 for DeepSORT. These results confirm that our approach effectively balances speed, accuracy, and identity stability, making it well-suited for people counting and surveillance applications.