Related Experiment Video
Updated: Mar 7, 2026
06:25
Author Spotlight: Assessment of Visual Acuity in Central Vision Loss Through Motion-Based Peripheral Vision Testing
Published on: February 23, 2024
1.2K
An Effective 3D Shape Descriptor for Object Recognition with RGB-D Sensors
Zhong Liu1, Changchen Zhao2, Xingming Wu3
1School of Automation Science and Electrical Engineering, Beihang University, Beijing 100191, China. lzpro@126.com.
Sensors (Basel, Switzerland)
|March 2, 2017
Summary
This study introduces a hybrid shape descriptor using RGB-D sensors for object recognition. The novel descriptor combines 2D and 3D features, outperforming existing methods in category and instance recognition tasks.
Area of Science:
- Computer Vision
- 3D Object Recognition
- Sensor Data Analysis
Background:
- RGB-D sensors are crucial in computer vision and graphics.
- Effective shape descriptors are vital for enhancing recognition performance.
- Analyzing multi-modality data from RGB-D sensors is key for robust object recognition.
Purpose of the Study:
- To propose a hybrid shape descriptor for object recognition using RGB-D sensor data.
- To evaluate the recognition performance of the proposed descriptor for both category and instance recognition.
- To analyze the contribution of individual features and computational complexity.
Main Methods:
- Extracted five 2D shape features from contour-based images.
- Extracted five 3D shape features from point cloud data.
- Combined 2D and 3D features into a hybrid descriptor for object representation.
Main Results:
- The hybrid shape descriptor demonstrated superior performance compared to common global-to-global descriptors.
- The descriptor achieved comparable accuracy to state-of-the-art partial-to-global descriptors.
- Analysis confirmed the significant contribution of partial features and manageable computational complexity.
Conclusions:
- The proposed hybrid shape descriptor effectively captures object characteristics for recognition.
- These shape features are valuable for object recognition and can enhance other feature-based methods.
- The approach offers a robust solution for object recognition using RGB-D data.
