Objectives Multi-object tracking (MOT) is a pivotal research area within the computer vision domain. Despite significant strides in MOT research, the field continues to grapple with formidable challenges: Indistinct appearance attributes of objects,objects exhibit irregular motion, anomalies in tracking arising from rigid trajectory lifecycle management strategies. These elements substantially undermine the precision and robustness of multi-object tracking endeavors.

Methods In response to these challenges, we present an advanced multi-object tracking algorithm that integrates dilatation intersection over union (DIOU) matching with an adaptive trajectory management approach. Initially, we introduce a metric based on a refined DIOU area for the primary matching between active trajectories and high-confidence detections, thereby improving the direct matching performance for high-quality detection boxes. Subsequently, for the re-matching of active trajectories with low-confidence detections, we implement a metric centered on a moderately dilated DIOU area, enhancing the tracking continuity of these detections. Furthermore, for reconnecting inactive trajectories with unmatched high-confidence detections, we employ a metric utilizing an extensively dilated DIOU area to bolster the probability of reactivating dormant trajectories. Lastly, an adaptive trajectory management strategy predicated on detection confidence scores is deployed to dynamically modulate the lifespan of trajectories, thereby mitigating the incidence of tracking anomalies and identity switches induced by occlusions and misidentifications.

Results (1) The application of the DIOU-based matching framework has yielded 5.4% increase in HOTA（higher order tracking accuracy） and a 1.5% increase in MOTA（multiple object tracking accuracy） on the DanceTrack dataset, corroborating the method's efficacy in densely populated scenes and complex motion environments. (2) The implementation of the adaptive trajectory management module has further resulted in 4.6% rise in HOTA, 0.8% elevation in MOTA, and 2.1% improvement in IDF1（identification F-score） on the DanceTrack dataset, demonstrating its capacity to efficiently counteract the limitations of fixed lifecycle sensitivities to false detections and missed detections.

Conclusions Although the refinement of data association and trajectory management strategies has led to a surge in tracking accuracy, the layering of multiple strategies has introduced a trade-off with computational efficiency, curtailing the peak performance of the tracking system.