TileTracker: Tracking Based Vector HD Mapping using Top-Down Road Images

TL;DR

A tracking-based HD mapping algorithm for top-down road images, referred to as tile images, is proposed, modified the BEVFormer layers to generate BEV masks from tile images, which are then used by the model to generate divider and boundary lines.

Abstract

In this paper, we propose a tracking-based HD mapping algorithm for top-down road images, referred to as tile images. While HD maps traditionally rely on perspective camera images, our approach shows that tile images can also be effectively utilized, offering valuable contributions to this research area as it can be start of a new path in HD mapping algorithms. We modified the BEVFormer layers to generate BEV masks from tile images, which are then used by the model to generate divider and boundary lines. Our model was tested with both color and intensity images, and we present quantitative and qualitative results to demonstrate its performance.

Figures (6)

• Figure 1: Main TileTracker model structure. Green and blue boxes show BEV and vector modules, respectively.
• Figure 2: An example of intensity (left) and color (right) images is shown. These images are generated from 3D colored lidar point clouds and displayed as 2D top-down BEV images, with intensity represented in grayscale and color images in their original colors.
• Figure 3: Unmerged and merged predictions (top left and right) as well as Unmerged and merged GTs (bottom left and right) for a curved path
• Figure 4: Unmerged and merged predictions (top left and right) as well as Unmerged and merged GTs (bottom left and right) for a path containing straight paths and turns
• Figure 5: Unmerged and merged predictions (top left and right) as well as Unmerged and merged GTs (bottom left and right) for a curved path