Systematic Field-of-view estimation from arbitrary viewpoints using LiDAR point clouds in community roads

Unsignalized intersections in Japanese community roads tend to have poor visibility due to occlusions such as walls or buildings, potentially leading to rush-out collisions. In particular, vulnerable road users (VRUs) such as pedestrians and cyclists are prone to serious injuries. Therefore, collision avoidance systems (CAS) designed to protect VRUs are essential. With the goal of developing such CAS, existing research aims to develop methods of evaluating the collision risk at intersections. A key factor of risk evaluation is evaluating the visibility under interception. Existing visibility evaluation methods have targeted vehicles equipped with on-board sensors. However, such methods are not applicable to VRUs, who typically do not possess such sensors. Therefore, we propose a novel CAS framework, which evaluates the visibility from arbitrary viewpoints beforehand and provides traffic participants with the data. Figure 1 shows the flow of the proposed framework. The proposed framework enables vulnerable traffic participants to obtain an evaluation of the visibility of the intersection from the real-time viewpoint by utilizing the GPS location data embedded in smartphones or other devices. The objective of this research is to develop a method for generating a digital map that records the visibility of arbitrary on-road viewpoints.

Fig. 1  Flow of proposed CAS framework

　In the proposed method, LiDAR point cloud data around intersections are linked with digital maps recording the location of intersections and roads. This enables the system to calculate the visibility of intersections from arbitrary viewpoints on the simulator. We used Field-of-View (FOV) as a metric of visibility, defined as the yaw angle of the visible area of the intersection. FOV is calculated by splitting the area around the viewpoint with a constant degree and checking the presence of points in each area. FOV was calculated for the left and right sides with respect to the longitudinal direction of the road. Figure 2 shows an overview of the calculation of FOV at a crossroad. The proposed method calculates the left and right FOV from equally spaced viewpoints on the road and preserves the result as a FOV map. A comparison between the FOV map and the actual environment confirmed that the proposed method calculates FOV corresponding to the visibility of the actual environment and the position on the road.

Fig. 2  Schematic of FOV calculation at a crossroad

The proposed FOV map generation system can be implemented in society at low cost by utilizing the driving data from delivery robots or intelligent mobilities, which are expected to become widespread in the future. Additionally, FOV maps have only  a few thousandths of of data volume of point cloud data, which is suitable for high speed processing on smartphones.

Next, we constructed a method to evaluate the risk from VRUs’ viewpoints using the FOV map. The proposed method uses the GPS installed in the VRUs’ smartphones to extract the FOV of the current location from the FOV map. However, the low precision of the GPS can lead to extracting the FOV of points away from the current location, resulting in inaccurate risk evaluation. Therefore, this method firstly assumes the current location as a 2D normal distribution with the low-level GPS measurement point at the center, to take into consideration the low precision of the GPS. Secondly, we set referenced points equally spaced on the road and evaluate the intersection risk for each point. Thirdly, we calculate the weighted sum of the risk for all points in which the existence probability at the point is the weight. Figure 3 shows the schematic of the proposed method.

Fig. 3  Schematic of risk evaluation method

To evaluate the risk evaluation system based on the proposed method, we compared it with a deterministic risk evaluation system in which the position estimation error is not taken into consideration. The results show that the proposed system predicted the risk at the current location with better precision, which proves the usefulness of the proposed method. As a future work, we would implement a CAS utilizing VRUs’ smartphones or other devices. 

Related Presentation

Hori, K., Tanaka, T., Watanabe, K., Nishi, M., Ogino, S. and Ito, T., Initial Study on Systematic Field-of-View Estimation of Unsignalized Intersections from Arbitrary Viewpoints Using LiDAR Point Clouds, in Proceedings of the ITS Japan 22nd ITS Symposium, Dec. 2024 

Acknowledgements

This projects was subsidized by JKA through its promotion funds from KEIRIN RACE.