Scalable Radar-based Roadside Perception: Self-localization and Occupancy Heat Map for Traffic Analysis

TL;DR

This work proposes a self-localization approach that outperforms other map-based radar localization methods, especially for the orientation estimation and projects radar sensor data onto a city-scale laser scan and generates a scalable occupancy heat map as a traffic analysis tool.

Abstract

4D mmWave radar sensors are suitable for roadside perception in city-scale Intelligent Transportation Systems (ITS) due to their long sensing range, weatherproof functionality, simple mechanical design, and low manufacturing cost. In this work, we investigate radar-based ITS for scalable traffic analysis. Localization of these radar sensors at city scale is a fundamental task in ITS. For flexible sensor setups, it requires even more effort. To address this task, we propose a self-localization approach that matches two descriptions of the "road": the one from the geometry of the motion trajectories of cumulatively observed vehicles, and the other one from the aerial laser scan. An Iterative Closest Point (ICP) algorithm is used to register the motion trajectory in the road section of the laser scan. The resulting estimate of the transformation matrix represents the sensor pose in a global reference frame. We evaluate the results and show that it outperforms other map-based radar localization methods, especially for the orientation estimation. Beyond the localization result, we project radar sensor data onto a city-scale laser scan and generate a scalable occupancy heat map as a traffic analysis tool. This is demonstrated using two radar sensors monitoring an urban area in the real world.

Figures (7)

• Figure 1: Overview of the scalable radar-based ITS. (a) Flexibly deployable sensor setups. (b) Self-localization based on sensor data and aerial laser scan data. The upper part of the figure illustrates the processing steps to extract the road section from the aerial laser scan with the help of a vectorized map. The lower part of the figure illustrates the description of the road section using the radar point cloud data. The voxelized road section from the aerial laser scan and the accumulated radar data are the input to the ICP registration algorithm. (c) Projection of the radar sensor data onto city scale laser scan. (d) Occupancy heat map as a tool for traffic analysis. Darker areas indicate high utilisation of the lanelet sector.
• Figure 2: The qualitative evaluation of the localization result is shown in the left figure. Dark blue points are the radar point clouds. Blue to green points are aerial laser scan points. The aligned dark blue clusters exactly overlap the tree canopy (marked by the rectangles). The right image from Google Earth is used as a reference for the reader to understand the scene.
• Figure 3: Quantitative evaluation of the localization. 50 initial poses are randomly selected around the ground truth value of the sensor setup for the ICP process. The error between the ICP result and the ground truth is given in the table.
• Figure 4: Alignment of the radar point cloud to the road. The lane description is manually created based on an aerial image (dashed lines). The lane information is also used for the masking process in Fig. \ref{['Figure:01_Concept']}(b). The upper zoomed image shows that the green area on the road corresponds to the untraveled area in the trajectory point cloud. The lower image shows that the traces of the left-turning vehicles match the lane description.
• Figure 5: Test points in the Carla map Town 10. The points and the line segments show the position and orientation of the radar sensors in test DY. In test DO the position is maintained but the orientation is more flexible and not shown in the figure.