<p>Multi-object tracking (MOT) is a critical task in computer vision, aiming to track multiple objects in video sequences while accurately identifying and localizing their identities and positions. However, the accuracy of tracking is still affected by factors such as object occlusion and tiny object. In this study, we propose a multi-object tracking method based on spatial super-resolution and spatiotemporal convolutional self-attention. First, to solve the problem of missed detection of small objects, we utilize spatial super-resolution algorithm, extracting more detailed information about object features. Then, to avoid false positive identification and ID switches due to object occlusion, we empower the DETR model by adding spatiotemporal convolutional self-attention layers to leverage the locality information between patches. Furthermore, we design a grid-based object query to the position encoding of the model, which makes up for the lack of explicit physical meanings in DETR. Experimental results on the MOT17 and MOT20 datasets demonstrate significant performance improvements with our proposed method.</p>