# Visual SLAM: What are the Current Trends and What to Expect?

**Authors:** Ali Tourani, Hriday Bavle, Jose Luis Sanchez-Lopez, Holger Voos  
**arXiv:** [2210.10491](https://arxiv.org/abs/2210.10491)  
**Date:** 2022-10-19  
**Summary page:** https://www.summarizepaper.com/en/paper/2210.10491v2/

## Key points

- Vision-based sensors are popular in SLAM systems for their performance, accuracy, and efficiency gains
- VSLAM methods outperform traditional methods by using cameras for pose estimation and map generation
- Challenges in VSLAM include loop closure detection optimization to prevent drift errors in scenarios with few feature points
- Object detection or line features can complement VSLAM methods to address challenges
- Recent advancements focus on improving image retrieval through visual vocabulary training and local feature aggregation

## Summary

, , , , 

Vision-based sensors have become increasingly popular in Simultaneous Localization and Mapping (SLAM) systems due to their significant performance, accuracy, and efficiency gains. Visual Simultaneous Localization and Mapping (VSLAM) methods utilize cameras for pose estimation and map generation, outperforming traditional methods that rely on a single sensor like Lidar. VSLAM approaches employ various camera types, datasets, environments, algorithms, and methodologies to enhance environmental understanding. One of the primary challenges in VSLAM is loop closure detection optimization to prevent drift errors in challenging scenarios with few salient feature points. Complementary scene-understanding methods like object detection or line features can help address this issue. Recent advancements in VSLAM systems have focused on improving image retrieval through visual vocabulary training and aggregation of local features. The paper categorizes recent works in VSLAM based on experimental environment, novelty domain, object detection/tracking algorithms, semantic level viability, performance metrics, etc. It also reviews critical contributions, existing drawbacks/challenges, future improvements suggested by authors, and trends in VSLAM systems. The discussion includes open issues that researchers are likely to investigate further. Notable examples of VSLAM systems include an indirect system using Occupancy Grid Mapping for high-accuracy localization and user interaction in GPS-denied conditions. Another method utilizes planes for tracking and graph optimization with real-time performance tested on indoor/outdoor datasets but limited support for geometric shapes. Analyzing current trends in VSLAM reveals that most proposed systems are standalone applications implementing localization and mapping from scratch. Improving Visual Odometry module emerges as a top objective among VSLAM applications. The visualization of processed data highlights the dominance of standalone applications over base platforms like ORB-SLAM 2.0 or ORB-SLAM for creating new frameworks. In conclusion, the survey provides insights into recent advancements in VSLAM systems while addressing challenges such as loop closure detection optimization and working in challenging scenarios with limited feature points. The discussion on current trends sheds light on the prevalent objectives pursued by researchers in the field of Visual Simultaneous Localization and Mapping.