At present, video data shows an explosive growth trend, with large redundancy and a lack of geographic information. When a user is not familiar with the monitoring area, it is difficult for the user to locate and measure dynamic objects in surveillance videos. In this paper, we aim to obtain dynamic objects’ 3D trajectories and visualize them in a 3D model. YOLOX was the detector and BYTE was used as the data association method. For a calibrated camera, we built a mapping model constrained by a plane. This model connects the image space with the real world. In the experiment, we used positions measured by a total station as the true values. Compared them with the positions calculated by the mapping model. The root mean square error was 0.159 m. The positioning accuracy is high. 3D geographic trajectories were generated by this mapping model and visualized in a 3D geographical scene.