Efficient Map Sparsification Based on 2D and 3D Discretized Grids

Xiaoyu Zhang,Yun-Hui Liu

2023-03-20

Abstract:Localization in a pre-built map is a basic technique for robot autonomous navigation. Existing mapping and localization methods commonly work well in small-scale environments. As a map grows larger, however, more memory is required and localization becomes inefficient. To solve these problems, map sparsification becomes a practical necessity to acquire a subset of the original map for localization. Previous map sparsification methods add a quadratic term in mixed-integer programming to enforce a uniform distribution of selected landmarks, which requires high memory capacity and heavy computation. In this paper, we formulate map sparsification in an efficient linear form and select uniformly distributed landmarks based on 2D discretized grids. Furthermore, to reduce the influence of different spatial distributions between the mapping and query sequences, which is not considered in previous methods, we also introduce a space constraint term based on 3D discretized grids. The exhaustive experiments in different datasets demonstrate the superiority of the proposed methods in both efficiency and localization performance. The relevant codes will be released at <a class="link-external link-https" href="https://github.com/fishmarch/SLAM_Map_Compression" rel="external noopener nofollow">this https URL</a>.

Robotics,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the memory consumption and computational efficiency issues faced by map construction and localization techniques when robots perform autonomous navigation in large - scale environments. As the scale of the map increases, the resources required to store and process these maps also increase, which leads to an inefficient localization process. To address these problems, the paper proposes an efficient map sparsification method, aiming to select a subset from the original map for localization while maintaining localization performance.

Specifically, the paper mainly focuses on the following two aspects of problems:

1. **Improving computational efficiency**: Existing map sparsification methods usually use Mixed - Integer Quadratic Programming (MIQP) to ensure the uniform distribution of selected landmarks, but this requires a large amount of memory and computational resources. For this reason, the paper proposes a map sparsification method in a linear form. By selecting uniformly distributed landmarks through a 2D discrete grid, the map can be sparsified while maintaining computational efficiency.

2. **Considering landmark visibility**: Previous methods assume that the spatial distribution of the query sequence is similar to that of the mapping sequence, but this assumption may not hold in practical applications. Therefore, the paper also introduces a spatial constraint term based on a 3D discrete grid to reduce the impact of different spatial distributions on localization performance and ensure that query images from the entire 3D space can be successfully localized.

Through the above methods, the paper aims to provide a map sparsification solution that is both efficient and can adapt to different query perspectives, thereby enhancing the autonomous navigation ability of robots in large - scale environments.