Fall Detection from Indoor Videos using MediaPipe and Handcrafted Feature

2025-03-03

AI Summary

To address the need for real-time, in-home fall detection among older adults, this paper proposes a non-intrusive, low-cost vision-based fall detection method. Methodologically, it leverages MediaPipe to extract human skeletal keypoints and integrates domain-informed, lightweight handcrafted features—including trunk inclination angle, abrupt lower-limb velocity changes, and abnormal support duration—classified via SVM or Random Forest. The key contribution lies in the first deep coupling of MediaPipe skeletal sequences with interpretable kinematic and geometric features, establishing a fall discrimination paradigm that balances generalizability and interpretability. Evaluated on the UR Fall dataset, the method achieves 98.2% accuracy, substantially outperforming existing vision-based approaches. Moreover, it demonstrates robustness across diverse age groups, genders, and environmental settings.

Technology Category

Computer Vision: Low Level & Physics-based VisionIntelligent Robots: Multimodal Perception & Sensor FusionKnowledge Representation and Reasoning: Common-Sense Reasoning

Application Category

Web Mining and Content Analysis: Robustness and generalizability of Web mining methodsResponsible Web: Human-perceived consequences of algorithmic deployment on the webSecurity and Privacy: Large-scale security measurements

Abstract

Falls are a common cause of fatal injuries and hospitalization. However, having fall detection on person, in particular for senior citizens can prove to be critical. Presently,there are handheld, ambient detector and vision-based detection techniques being utilized for fall detection. However, the approaches have issues with accuracy and cost. In this regard, in this research, an approach is proposed to detect falls in indoor environments utilizing the handcrafted features extracted from human body skeleton. The human body skeleton is formed using MediaPipe framework. Results on UR Fall detection show the superiority of our model, capable of detecting falls correctly in a wide number of settings involving people belonging to different ages and genders. This proposed model using MediaPipe for fall classification in daily activities achieves significant accuracy compare to the present existing approaches.

Problem

• Detects falls in indoor environments using MediaPipe and handcrafted features.
• Addresses accuracy and cost issues in existing fall detection methods.
• Improves fall detection for diverse age and gender groups.

Innovation

• Uses MediaPipe for human body skeleton extraction
• Employs handcrafted features for fall detection
• Achieves high accuracy across diverse settings