Unilever Smart Staircase Safety
Staircase safety monitoring for handrail use and phone usage using YOLO person detection, MediaPipe Pose, dynamic handrail regions, wrist geometry, temporal stability, and evidence delivery.
Workflow at a glance
Locate people
Detect people within staircase regions.
Estimate pose
Locate body landmarks for behavior checks.
Check geometry
Assess handrail contact and phone-use cues.
Confirm and review
Consolidate observations and deliver evidence.
Project facts & reported results
Implementation fact
Implementation fact
Implementation fact
Project Overview
Problem Statement
Unsafe stair behavior is brief and posture-dependent, so reliable detection requires pose geometry and confirmation across time rather than single-frame object detection.
Approach & Methodology
Implemented MediaPipe Pose for real-time human pose estimation. Built risk assessment model analyzing gait patterns, handrail usage, and posture. Created alert system for unsafe behaviors (running, improper posture, phone usage). Integrated with existing CCTV infrastructure.
Outcome & Scope
Provides reviewable alerts for defined safety behaviors while using temporal checks to reduce transient pose errors.
Technical Challenges & Solutions
Pose Landmark Segmentation Failures
MediaPipe Pose landmarks failed when multiple people overlapped on stairs, causing incorrect railing detection and missed safety violations.
Implemented YOLO11-based person detection as preprocessing to segment individuals before pose estimation. Each person processed independently with dedicated landmark model.
Multi-Person Simultaneous Tracking
System initially designed for single person, failed when multiple people were on stairs simultaneously due to landmark confusion.
Upgraded to multi-instance pose estimation pipeline. Each detected person gets unique tracking ID with DeepSORT for temporal consistency across frames.
System Architecture
Real-time pose-based safety monitoring system with multi-floor coverage
Key Components:
Manages camera feeds from multiple staircase locations with centralized processing
Detects and segments individual people on stairs for isolated pose analysis
Extracts 33 body landmarks per person for behavior analysis
Analyzes pose landmarks to detect railing grip, phone usage, and unsafe postures
Maintains consistent IDs across frames using DeepSORT for behavior history
Generates real-time audio warnings and dashboard notifications for unsafe behaviors
Records violations with video clips for safety training and compliance reporting
Key Features & Capabilities
Railing grip detection via pose landmarks
Phone usage detection while on stairs
Multi-person simultaneous tracking
Multi-floor coverage with centralized AI
Real-time alerts via web dashboard
Live video analysis with NVR storage