Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Analysis and Evaluation of Kinect-based Action Recognition Algorithms

Lei Wang

TL;DR

This work benchmarks four Kinect-based action-recognition methods (HON4D, HDG, HOPC, RBD) across five challenging datasets to understand robustness to viewpoint, noise, and occlusion. It includes an implemented and enhanced HDG pipeline and applies it to cross-view recognition on UWA3D Multiview, with systematic hyperparameter optimization. Key findings show that skeleton-based representations (RBD, HDG with joint features) can outperform depth-only descriptors in cross-view tasks, while HOPC often excels in single-view scenarios; combining depth and skeleton information generally improves performance. The results highlight practical implications for robust action recognition and point toward CNN-based extensions for real-time, viewpoint-agnostic systems in cluttered environments.

Abstract

Human action recognition still exists many challenging problems such as different viewpoints, occlusion, lighting conditions, human body size and the speed of action execution, although it has been widely used in different areas. To tackle these challenges, the Kinect depth sensor has been developed to record real time depth sequences, which are insensitive to the color of human clothes and illumination conditions. Many methods on recognizing human action have been reported in the literature such as HON4D, HOPC, RBD and HDG, which use the 4D surface normals, pointclouds, skeleton-based model and depth gradients respectively to capture discriminative information from depth videos or skeleton data. In this research project, the performance of four aforementioned algorithms will be analyzed and evaluated using five benchmark datasets, which cover challenging issues such as noise, change of viewpoints, background clutters and occlusions. We also implemented and improved the HDG algorithm, and applied it in cross-view action recognition using the UWA3D Multiview Activity dataset. Moreover, we used different combinations of individual feature vectors in HDG for performance evaluation. The experimental results show that our improvement of HDG outperforms other three state-of-the-art algorithms for cross-view action recognition.

Analysis and Evaluation of Kinect-based Action Recognition Algorithms

TL;DR

This work benchmarks four Kinect-based action-recognition methods (HON4D, HDG, HOPC, RBD) across five challenging datasets to understand robustness to viewpoint, noise, and occlusion. It includes an implemented and enhanced HDG pipeline and applies it to cross-view recognition on UWA3D Multiview, with systematic hyperparameter optimization. Key findings show that skeleton-based representations (RBD, HDG with joint features) can outperform depth-only descriptors in cross-view tasks, while HOPC often excels in single-view scenarios; combining depth and skeleton information generally improves performance. The results highlight practical implications for robust action recognition and point toward CNN-based extensions for real-time, viewpoint-agnostic systems in cluttered environments.

Abstract

Human action recognition still exists many challenging problems such as different viewpoints, occlusion, lighting conditions, human body size and the speed of action execution, although it has been widely used in different areas. To tackle these challenges, the Kinect depth sensor has been developed to record real time depth sequences, which are insensitive to the color of human clothes and illumination conditions. Many methods on recognizing human action have been reported in the literature such as HON4D, HOPC, RBD and HDG, which use the 4D surface normals, pointclouds, skeleton-based model and depth gradients respectively to capture discriminative information from depth videos or skeleton data. In this research project, the performance of four aforementioned algorithms will be analyzed and evaluated using five benchmark datasets, which cover challenging issues such as noise, change of viewpoints, background clutters and occlusions. We also implemented and improved the HDG algorithm, and applied it in cross-view action recognition using the UWA3D Multiview Activity dataset. Moreover, we used different combinations of individual feature vectors in HDG for performance evaluation. The experimental results show that our improvement of HDG outperforms other three state-of-the-art algorithms for cross-view action recognition.
Paper Structure (20 sections, 3 equations, 12 figures, 5 tables)

This paper contains 20 sections, 3 equations, 12 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Algorithms to be Analyzed and Evaluated
  4. The HON4D algorithm
  5. The HDG algorithm
  6. The HOPC algorithm
  7. The RBD algorithm
  8. Experimental Settings
  9. Benchmark Datasets
  10. Evaluation Settings
  11. Evaluation Measure
  12. Optimisation of Hyperparameters for HDG
  13. Results and Discussions
  14. Results and Discussions for the first 4 datasets
  15. MSRAction3D Dataset
  16. ...and 5 more sections

Figures (12)

  • Figure 1: Steps for the HON4D descriptor computation Oreifej2013.
  • Figure 2: The HDG for real-time action recognition Rahmani2014.
  • Figure 3: The training and testing phase using the RBD algorithm Vemulapalli2016.
  • Figure 5: Normalized feature importance of 13250 predictors with the original index preserved for the MSRAction3D dataset.
  • Figure 6: Average recognition accuracy for the MSRAction3D dataset using HDG-all features with different number of trees trained and different threshold factors for feature selection.
  • ...and 7 more figures