Opportunity
Fall risk increases significantly with age—61.8% among those aged 65-74 and 88.6% among those 75 and above. Falls cause severe injuries, fractures, financial burdens, and decreased self-care ability. Early identification of fall risk factors and effective prevention strategies are critically needed. Existing monitoring systems often use wearable devices (which elderly users may forget or refuse to wear) or intrusive cameras that raise serious privacy concerns. Many systems only detect falls after they happen, not before—alerting caregivers only after injury has occurred. There is a need for a non-intrusive, privacy-preserving system that can predict potential falls and hazardous activities before accidents happen, enabling proactive intervention.
Technology
This patent presents an accident prevention system that uses a standard camera module to capture images of a target user (e.g., elderly person) in a real-world environment (home, care facility). A pose estimation module analyzes these images to estimate the user's pose—but crucially, it filters out the user's identity by generating a graphical representation (a skeleton/stickman) instead of displaying the actual person. This protects privacy while preserving movement information.
An accident prediction module compares the estimated pose against predetermined pose references stored in a database. It predicts potential hazardous events prior to actual occurrence by analyzing multiple risk factors: the user's elevated level relative to ground (standing on chair, climbing ladder), presence of real-world objects (shelf, chair, table), interaction between user and objects (lifting heavy items, standing on unstable surfaces), and user velocity (running on slippery floors). The system uses a machine learning network model that continuously learns from captured data to improve prediction accuracy.
When a potential hazard is detected (e.g., standing on a chair, climbing a ladder, improper sitting, falling), the system sends an immediate alert to a remote caregiver's smartphone, displaying the estimated skeleton pose, timestamp, and hazard description. The system also records statistics of high-risk activities for long-term analysis and personalized fall prevention recommendations.
Advantages
- Predictive, Not Just Reactive: Detects potential hazards before accidents happen (e.g., standing on chair, climbing ladder), enabling proactive intervention.
- Privacy-Preserving: Displays only a stickman/skeleton representation; the user's actual image and identity are never shown or recorded.
- No Wearable Devices: Uses only a standard camera--no sensors or devices for the elderly to wear, remember, or charge.
- Real-Time Alerting: Immediately notifies remote caregivers via smartphone with timestamp and hazard description.
- Object & Activity Recognition: Detects household objects (chairs, shelves, tables) and analyzes user-object interactions (lifting, climbing, improper sitting).
- Continuous Learning: Machine learning model improves over time as more activity data is collected and analyzed.
- Long-Term Risk Analysis: Records statistics of high-risk activities, enabling personalized fall prevention recommendations.
Applications
- Elderly Home Care: Monitoring seniors living alone or in care facilities for fall risks without invading privacy.
- Rehabilitation Centers: Tracking patient movements to prevent falls during recovery from injuries or surgeries.
- Hospital Wards: Monitoring elderly or mobility-impaired patients to prevent bedside falls.
- Assisted Living Facilities: Centralized monitoring of multiple residents with privacy-preserving pose estimation.
- Remote Caregiving: Allowing family members to receive alerts about their elderly relatives' risky activities from anywhere.
