The heavy rains often have big impact on urban traffic, it causes a lot of water cumulating at road depression, which affects normal thoroughfare for vehicles and even brings about road congestion. In this paper, seed spread algorithm to determine the catchment area is improved through analyzing the topographic feature of the underpass bridge and its surrounding small area of city typical low-lying areas, and it solves the inaccurate problem of catchment area of small area. According to forecast information of meteorological department, accumulated water region is quickly simulated and the water depth is dynamically displayed through the analysis of the process of road depression flooded by heavy rains and the calculation of water-collecting amount and depth, which can provide early warning information. Through the secondary development of 3D GIS software-skyline, the early flood warning analysis and dynamic simulation of the urban rainstorm is implemented, which can provide decision support for traffic warning and emergency response.

The vehicle delay is a key problem in intersection signal control system. Because there is enough security space for pedestrians under urban overpasses, the traffic control approach of pedestrians crossing signalized intersection may be further improved. To the above problems, this paper shows two-phase control method in pedestrians crossing signalized intersection under urban overpasses. According to this method, the vehicle delay affected by pedestrians crossing the signalized intersection is studied. Meanwhile, the variation trend of vehicle delay under different road widths are also analyzed. At last, an overpass is conducted for example analysis, and the result shows that this approach is practical and feasible for the special case of pedestrians crossing signal control intersection under urban overpasses, it can effectively reduce the pedestrian and vehicle delay.

Extracting vehicles track at roundabout is the basis research for motion characteristics of vehicles. At present, most of the vehicle trajectory was obtained by using single data source. However, the precision of the method needs to be improved. In this paper, two cameras are set at two different directions shooting the same scene image to obtain image data. And then a new method to extracted vehicles track more accurately by integrate the dual data source organically was proposed. In the case analysis, through the contrastive analysis of 4 different curves, the result shows that the error of dual data detection is decreased by 2.7% than single data detection. The method to extract vehicles track by dual data source can provide the theoretical support for the analysis of traffic conditions at roundabout.

In order to simulate and analyze the rolling movement state of motorcycle in collision accidents, the trajectory reconstruction model was studied. The motorcycle’s centroid trajectory model during the process of rolling and the feature point trajectory model during the process of sliding are both constructed. Using the accidents reconstruction and analysis software named PC-Crash and the models constructed in this paper respectively, a real accident case was reconstructed and analyzed. The comparative study between the reconstruction analysis results and the marks remained in the accident scene was conducted. Research showed that, the analysis result reconstructed by the model was consistent with the actual accident scene. This model can be used to reconstruct and analyze the motorcycle’s rolling movement trajectory after collision occurring.

The video detection technology is a very active branch in the research field of machine vision, and the detection of vehicle trajectory by the stationary camera has become more and more important. However, there is little research on the measurement error of left-turn vehicles trajectory.This paper not only studies the relationship between the measurement error and longitudinal distance; but also relationship of measurement error and transverse distance. Their mathematical models are constructed and each of the variables in the models is analyzed. Then in the case analysis of certain intersection, the experimental results have shown that the proposed funtions can describe the relationship between measurement error and the parameters. It can provide the theoretical support for the measurement error problems of obtaining trajectory by video detection.