Related Experiment Video
Updated: Apr 15, 2026
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
1.2K
NeuroFusion-SLAM: A Deep Neural Network Framework for Real-Time Multi-Sensor SLAM.
Chenchen Yu1, Wei Wei1, Zhihong Cao1
1Shaanxi Key Laboratory for Network Computing and Security Technology, Xi'an University of Technology, Xi'an 710048, China.
Sensors (Basel, Switzerland)
|April 14, 2026
Summary
NeuroFusion-SLAM enhances visual SLAM (VSLAM) with a novel multi-sensor fusion framework. It achieves real-time performance and robustness by reducing computational costs and improving global consistency.
Area of Science:
- Robotics
- Computer Vision
- Artificial Intelligence
Background:
- Deep learning-based visual SLAM (VSLAM) offers high localization accuracy but suffers from computational cost and latency.
- Real-time deployment of VSLAM systems is hindered by these performance limitations.
Purpose of the Study:
- To develop an efficient and robust multi-sensor fusion framework for VSLAM.
- To address the computational bottlenecks of deep learning-based VSLAM systems.
Main Methods:
- Implemented NeuroFusion-SLAM, a novel multi-sensor fusion framework.
- Incorporated depthwise separable convolution to reduce model parameters and training time.
- Introduced a global edge optimization strategy using sliding window optimization and factor graphs.
Main Results:
- Achieved approximately 40% reduction in model parameters and 49% reduction in training time.
- Demonstrated real-time performance with an average latency of 30.4 ms per frame.
- Outperformed ORB-SLAM2 by 3x and VINS-Mono by 4x in speed while maintaining localization accuracy.
Conclusions:
- NeuroFusion-SLAM significantly improves computational efficiency and real-time inference performance.
- The proposed global edge optimization enhances the global consistency of VSLAM systems.
- The framework offers a robust and efficient solution for large-scale VSLAM applications.