Multi-object tracking aims to track multiple objects across consecutive frames in a video, assigning a unique classifier to each object. However, issues such as occlusions, directional changes, or shape alterations can cause appearance variations, leading to detection and matching problems that in turn result in frequent ID switches. To solve these issues, this paper proposes a two-stage multi-object tracking framework based on a spatial and temporal fusion algorithm. First, the video frames are processed by a detector to identify objects and form rectangular detection areas. Meanwhile, an estimator predicts the target rectangular areas in the next frame. Then, we extract the optical flow of the target pixels within the detection and prediction areas, and then a temporal information model is established by calculating the average of the target pixels’ optical flow. Afterward, we present a spatial information model using the R-IoU (Reverse of Intersection over Union) between the detection and prediction areas. This spatial and temporal information is combined with weighted matrix fusion, which achieves the feature matching and association task. Finally, we implement a two-stage association multi-object tracking model using the mentioned fusion algorithm. Experiments on the MOTChallenge dataset using the official detector show that our two-stage multi-object tracking method based on the spatial and temporal fusion algorithm is robust in handling occlusions and ID switch issues. As of the submission of this paper, the proposed method has achieved the top ranking in the MOT17 benchmark when evaluated with the official detector.

Open-world person re-identification (ReID) faces severe spatially asymmetric interferences, such as partial occlusion and illumination distortion. Existing models adopting static weight fusion irreversibly corrupt identity representations when processing localized noise. To address this, we propose a Scene-Prompt-Driven Dynamic Routing Expert Network (SPDR-Net). Guided by spatial topological prior constraints, SPDR-Net treats person semantic parts as independent local experts. First, a semantic expert module decouples and refines local features using human spatial topology priors. Second, a Prompt-Guided Dynamic Routing (PGDR) network extracts high-level scene contexts to dynamically evaluate each expert’s reliability, assigning routing weights to attenuate noise propagation. Finally, a global–local fusion module superimposes high-purity local features onto a global identity anchor, maintaining topological integrity to generate a unified descriptor for robust person re-identification. Extensive experiments on seven public benchmarks and a newly constructed real-world street dataset (SD-ReID) demonstrate that SPDR-Net achieves highly competitive performance against state-of-the-art methods and exhibits superior robustness under severe spatial-asymmetric interferences. Furthermore, end-to-end multi-camera closed-loop tests verify its robust decision-making capability and high engineering application value in real-world security systems.