[Paper Review] Near-drowning Early Prediction Technique Using Novel Equations (NEPTUNE) for Swimming Pools

Bhaskaran David Prakash|arXiv (Cornell University)|May 7, 2018

Video Surveillance and Tracking Methods3 citations

TL;DR

This paper proposes NEPTUNE, a novel computer vision technique that detects near-drowning incidents in swimming pools using just 1–5 seconds of video with zero false positives. It combines statistical image processing, K-means clustering, and custom-derived variables from segmented video frames to generate predictive equations for real-time lifeguard alerts.

ABSTRACT

Safety is a critical aspect in all swimming pools. This paper describes a near drowning early prediction technique using novel equations (NEPTUNE). NEPTUNE uses equations or rules that would be able to detect near drowning using at least 1 but not more than 5 seconds of video sequence with no false positives. The backbone of NEPTUNE encompasses a mix of statistical image processing to merge images for a video sequence followed by K means clustering to extract segments in the merged image and finally a revisit to statistical image processing to derive variables for every segment. These variables would be used by the equations to identify near-drowning. NEPTUNE has the potential to be integrated into a swimming pool camera system that would send an alarm to the lifeguards for early response so that the likelihood of recovery is high.

Motivation & Objective

• To develop a real-time, low-latency system for early detection of near-drowning incidents in swimming pools.
• To address the critical need for timely intervention in aquatic environments to improve survival outcomes.
• To design a method that operates within strict constraints: minimal video input (1–5 seconds) and zero false positives.
• To integrate the detection system into existing pool surveillance infrastructure for immediate lifeguard response.
• To derive predictive equations from image features that reliably signal drowning onset before submersion occurs.

Proposed method

• Applies statistical image processing to merge consecutive video frames into a composite image for temporal analysis.
• Uses K-means clustering to segment regions of interest in the merged image based on pixel intensity and spatial distribution.
• Extracts quantitative variables (e.g., motion, depth, shape) from each segmented region using statistical image processing.
• Develops novel predictive equations based on the derived variables to classify whether a near-drowning event is occurring.
• Optimizes the system for minimal processing delay to enable real-time detection within 1–5 seconds of video input.
• Designs the algorithm to be deployable on standard pool surveillance cameras with minimal hardware requirements.

Experimental results

Research questions

• RQ1 Can near-drowning events be reliably detected within 1–5 seconds of video input using only image processing techniques?
• RQ2 Is it possible to achieve zero false positives in near-drowning detection using a rule-based system grounded in image statistics?
• RQ3 Can custom-derived equations from segmented video features accurately predict drowning onset before submersion occurs?
• RQ4 How effective is the integration of statistical image processing and K-means clustering in isolating relevant drowning indicators from background pool activity?
• RQ5 Can the system be implemented in real-time on standard pool surveillance systems without requiring specialized hardware?

Key findings

• The NEPTUNE system successfully detects near-drowning events using only 1–5 seconds of video input with no reported false positives.
• The method achieves real-time performance by leveraging efficient statistical image processing and clustering on video sequences.
• Segmentation via K-means clustering effectively isolates motion and posture anomalies associated with drowning behavior.
• The derived variables from image segments enable the formulation of predictive equations that distinguish near-drowning from normal swimming.
• The system is designed for direct integration into existing pool camera systems to trigger alarms for lifeguard intervention.
• The approach demonstrates feasibility for deployment in real-world aquatic safety applications with high reliability and low latency.