Using Detection, Tracking and Prediction in Visual SLAM to Achieve Real-time Semantic Mapping of Dynamic Scenarios

TL;DR

This paper introduces RDS-SLAM, a lightweight real-time semantic SLAM system that effectively tracks and maps dynamic environments using only a standard CPU, enhancing robotic perception and interaction capabilities.

Contribution

The paper presents RDS-SLAM, a novel lightweight SLAM system with integrated object detection, tracking, and semantic mapping, optimized for real-time performance on a single CPU.

Findings

Runs at 30.3 ms per frame on an Intel Core i7 CPU

Achieves accuracy comparable to state-of-the-art systems

Effectively handles dynamic scenarios in real-time

Abstract

In this paper, we propose a lightweight system, RDS-SLAM, based on ORB-SLAM2, which can accurately estimate poses and build semantic maps at object level for dynamic scenarios in real time using only one commonly used Intel Core i7 CPU. In RDS-SLAM, three major improvements, as well as major architectural modifications, are proposed to overcome the limitations of ORB-SLAM2. Firstly, it adopts a lightweight object detection neural network in key frames. Secondly, an efficient tracking and prediction mechanism is embedded into the system to remove the feature points belonging to movable objects in all incoming frames. Thirdly, a semantic octree map is built by probabilistic fusion of detection and tracking results, which enables a robot to maintain a semantic description at object level for potential interactions in dynamic scenarios. We evaluate RDS-SLAM in TUM RGB-D dataset, and experimental results show that RDS-SLAM can run with 30.3 ms per frame in dynamic scenarios using only an Intel Core i7 CPU, and achieves comparable accuracy compared with the state-of-the-art SLAM systems which heavily rely on both Intel Core i7 CPUs and powerful GPUs.