UAV 3D reconstruction technology provides explorable new ways to digitize traffic accident scenes. In this paper, a parallelized incremental UAV image 3D reconstruction algorithm (SFM) is proposed. Combining vocabulary tree retrieval and GPS constraints to optimize beam method leveling (BA), it significantly improves the reconstruction accuracy and speed of the accident scene. A traffic accident simulation system based on virtual reality technology is designed to realize pre-collision, post-collision, and initial speed inversion and multi-morphology accident simulation. The results show that: in the multi-group ranging experiments, the absolute error value of UAV ranging in this paper is 0.22 at maximum, and the relative error is only 0.43% at maximum, which is smaller than 5.30 and 8.76% of the traditional method. Combined with the image of 4.5m+25m aerial height to complete the modeling, the absolute error in x and y direction is only 0.00m and 0.58m, and the relative error is 0.125 and 0.250. Using the model to restore the situation at the time of the accident, the change of acceleration at the time of the three head injuries can be accurately calculated.
