Related Experiment Video
Updated: Jun 5, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
468
Exploiting the features of deep residual network with SVM classifier for human posture recognition.
Irfan Kareem1, Syed Farooq Ali2, Muhammad Bilal3,4
1Department of Mathematics and Computer Science, University of Calabria, Rende, Italy.
Plos One
|December 5, 2024
Summary
This study introduces a hybrid deep learning model for fall detection, achieving high accuracy by combining ResNet-50 features with Support Vector Machine classification. The novel approach significantly improves fall detection rates in realistic conditions.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Biomedical Engineering
Background:
- Human posture recognition has advanced significantly, with deep learning methods showing promise.
- Existing fall detection systems face challenges in realistic environments with variations in lighting, camouflage, and occlusion.
Purpose of the Study:
- To analyze the performance of a hybrid deep learning architecture for enhanced fall detection.
- To evaluate the fusion of deep features from Residual Network (ResNet-50) with Support Vector Machine (SVM) classification.
Main Methods:
- A hybrid deep learning model was developed, integrating ResNet-50 for feature extraction and SVM for classification.
- The model's performance was evaluated on Multi-Camera Fall (MCF), UR Fall detection (URFD), and UP-Fall detection (UPFD) datasets.
- Comparative experiments were conducted against six state-of-the-art deep learning networks and various classifiers (Naive Bayes, Decision Tree, Random Forest, KNN, AdaBoost, MLP).
Main Results:
- The proposed hybrid approach achieved superior accuracy, reaching 98.82% (MCF), 97.95% (URFD), and 99.98% (UPFD), outperforming existing methods.
- 100% accuracy was attained on the UPFD two-posture task.
- The model demonstrated high performance in realistic conditions (camouflage, occlusion, lighting variations) and outperformed other networks in accuracy and time efficiency.
- SVM classifier outperformed or matched other classifiers when integrated into the hybrid model.
Conclusions:
- The hybrid deep learning architecture combining ResNet-50 and SVM offers a highly accurate and efficient solution for fall detection.
- The proposed method shows robustness in challenging, real-world scenarios, marking a significant advancement in human posture recognition for safety applications.
