Intelligent Mobility Laboratory (Takuma Ito Laboratory) Department of Mechanical Engineering, Graduate School of Engineering, The University of Tokyo

Position estimation method for traffic participants using statistical information

Traffic accidents on community roads are a major social problem in Japan. It is difficult to prevent crossing accidents on community roads with poor visibility by using onboard sensors alone. Therefore, new countermeasures are required to prevent traffic accidents. In general, an effective countermeasure is to obtain information from roadside sensors and/or smartphones of traffic participants so that intelligent vehicles can recognize traffic participants in blind areas. However, owing to cost constraints, it is impractical to place roadside sensors so densely that the system can always observe traffic participants. Additionally, not all traffic participants are willing to provide information from their smartphones. Given this background, a position estimation method with limited real-time observation information is required. Therefore, we assume the situation shown in Fig.1, in which a cyclist observed by the roadside sensor exits the roadside sensor’s observation range and approaches the next intersection. In such a situation, we propose a method to continue position estimation after the traffic participant exits the observation range with a small uncertainty. To be more precise, we propose a method to generate virtual observations from statistical information and utilize virtual observations to compensate for the lack of real-time observation information. By using the virtual observation generated from statistical information, the proposed method can estimate the traffic cyclist’s position appropriately with a small uncertainty.

Figure 1

Figure 2 shows the schematic of the proposed method. The proposed method estimates the cyclist’s position by integrating the observed information from the actual roadside sensor and virtual observations. Kalman filter, which is a method that can estimate states considering the uncertainty of observed information, is used as integration method. Kalman filter outputs the estimated states x and its uncertainty P. In the situation assumed in this study, the challenge is that real-time information is limited. The proposed method generates a virtual observation from statistical information that represents the movement characteristics of the traffic participants. By using such virtual observations, the method can compensate for the lack of real-time observation and estimate traffic participants’ positions with small uncertainty. Here, the virtual observation z and its uncertainty R are calculated from the cyclists’ average velocity and velocity range. In the calculation of z and R, we decide not to overtrust the virtual observation.

Figure 2

A simulation was conducted to evaluate the proposed method. In the simulation, the proposed method performed better than conventional methods. Furthermore, we conduct an experiment to validate the feasibility of the proposed method in an actual environment. From the experimental results, it is confirmed that the proposed method can estimate the cyclists’ positions appropriately with a small uncertainty. As for the detailed results, please confirm it in the journal paper. In our future work, we will develop a method to adapt virtual observations to differences in traffic participants’ movement characteristics. In addition, we plan to develop a method for constructing statistical information.