Video object detection is a challenging task in the field of computer vision. Due to the influence of varying scales and motion blur in video objects, as well as the insufficient feature aggregation in existing algorithms, the improvement of detection performance is constrained. To enhance the accuracy of video object detection, this study proposes a method based on Transformer. First, this study designs a spatio-temporal feature aggregation layer based on spatio-temporal deformable attention and integrates it into the Transformer encoder to achieve multi-scale spatio-temporal feature aggregation encoding between the target frame and reference frames. Second, a method to enhance queries through feature-query interaction is proposed. Specifically, the spatio-temporal features of reference frames are directly utilized in the Transformer decoder to augment the target queries, allowing them to fully learn global spatio-temporal information. The method fully leverages the advantages of the Transformer encoder-decoder architecture, which effectively enhances the spatio-temporal representation capabilities of both target frame features and queries. Experimental results on the ImageNet VID and UA-DETRAC datasets demonstrate that the proposed method achieves an average precision (mAP) of 86.0% on the ImageNet VID dataset and 90.9% on the UA-DETRAC dataset, respectively, using a ResNet-101 backbone network.