Related Experiment Video
Updated: Nov 7, 2025
08:05
Design and Analysis for Fall Detection System Simplification
Published on: April 6, 2020
10.9K
Real-time 2D-3D door detection and state classification on a low-power device
João Gaspar Ramôa1,2, Vasco Lopes1,2, Luís A Alexandre1,2
1NOVA LINCS, Costa da Caparica, Portugal.
SN Applied Sciences
|May 4, 2021
Summary
We developed three novel methods for classifying door states (open, closed, semi-open) to enhance robot navigation. These algorithms run in real-time on low-power devices like the Jetson Nano.
Area of Science:
- Robotics
- Computer Vision
- Artificial Intelligence
Background:
- Robot navigation in indoor environments requires accurate perception of surroundings.
- Distinguishing door states (open, closed, semi-open) is crucial for efficient path planning.
- Existing methods often lack real-time performance or are limited to specific hardware.
Purpose of the Study:
- To propose and evaluate three distinct methods for door state classification.
- To enable real-time door state recognition on low-power computing platforms.
- To develop versatile algorithms applicable beyond door detection.
Main Methods:
- Utilized 3D object classification (PointNet) and real-time semantic segmentation (FastFCN, FC-HarDNet, SegNet, BiSeNet).
- Employed object detection (DetectNet) and 2D classification networks (AlexNet, GoogleNet).
- Created a comprehensive 3D and RGB door dataset using a Realsense D435 camera.
Main Results:
- All proposed methods achieved real-time performance on a low-power Jetson Nano device.
- The algorithms successfully differentiated between open, closed, and semi-open doors.
- Accuracy and speed were analyzed for each method on the low-power computer.
Conclusions:
- Real-time door classification on low-power devices is feasible.
- The developed methods offer a flexible solution for robot navigation and other applications.
- The freely available dataset supports further research in indoor environment perception.
