# DyOb-SLAM : Dynamic Object Tracking SLAM System

**Authors:** Rushmian Annoy Wadud, Wei Sun  
**arXiv:** [2211.01941](https://arxiv.org/abs/2211.01941)  
**Date:** 2022-11-03  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2211.01941v1/

## Key points

- Simultaneous Localization and Mapping (SLAM) is a crucial process in robotics
- SLAM involves building a relationship between localization and mapping of the subject in its surrounding environment
- Various SLAM systems have been developed using different sensors
- Traditional SLAM processes lack consideration for dynamic objects in the mapping of the environment
- The authors propose a Dynamic Object Tracking SLAM (DyOb-SLAM) system to overcome this limitation
- DyOb-SLAM is a Visual SLAM system that localizes, maps static features, and tracks dynamic objects in each frame
- It utilizes a neural network and dense optical flow algorithm to differentiate between dynamic and static objects
- DyOb-SLAM creates two separate maps: one for static features represented by a sparse map obtained through Bundle Adjustment algorithm, and another for dynamic contents represented by a trajectory global map updated with time
- DyOb-SLAM enables estimation of object speeds by calculating both camera pose and object pose in real-time frame-to-frame tracking system
- Performance evaluation of DyOb-SLAM compared to VDO-SLAM includes measuring camera and object pose errors as well as object speed error
- The proposed system shows promising performance compared to existing methods, highlighting its potential for various robotic applications.

## Summary

Simultaneous Localization and Mapping (SLAM) is a crucial process in robotics that involves building a relationship between localization and mapping of the subject in its surrounding environment. Various SLAM systems have been developed to address this problem using different sensors. To overcome the limitation of traditional SLAM processes which lack consideration for dynamic objects in the mapping of the environment, the authors propose a Dynamic Object Tracking SLAM (DyOb-SLAM) system. DyOb-SLAM is a Visual SLAM system that not only localizes and maps static features but also tracks dynamic objects in each frame. This is achieved by utilizing a neural network and a dense optical flow algorithm to differentiate between dynamic and static objects in the environment. DyOb-SLAM creates two separate maps: one for static features represented by a sparse map obtained through Bundle Adjustment algorithm; and another for dynamic contents represented by a trajectory global map updated with time. The key advantage of DyOb-SLAM is its real-time frame-to-frame based dynamic object tracking system which enables estimation of object speeds by calculating both camera pose and object pose. The performance of DyOb-SLAM is evaluated by comparing it with VDO-SLAM, measuring camera and object pose errors as well as object speed error. The proposed system shows promising performance compared to existing methods, highlighting its potential for various robotic applications.