Most of the existing autonomous driving algorithms are based on on-board sensors, ignoring the study of traffic data set from roadside perspective. Diffusion model has been widely used in various visual tasks due to its unique generation mechanism, but its potential in discriminant tasks remains to be explored. In response to the above, an integrated method of object detection and pose estimation based on diffusion modeling (D-DetPose) is proposed. The problem is solved by converting object detection and pose estimation into a generative model on the position and size space of bounding boxes in the image, the training phase diffuses the truth-valued boxes to a random distribution, and the model learns to invert this noisy process, refining a set of randomly generated boxes to the output results of categories, object boxes and poses. Experiments on the roadside dataset Rope3D show that mAP reaches 72.10%, which is 24.51% higher than BevHeight, and 2.58%, 10.45% and 10.59% higher than BevHeight at 5m5° of car, truck and pedestrian, respectively.