Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

System and Method for Automated Lifeguard

IP.com Disclosure Number: IPCOM000240612D
Publication Date: 2015-Feb-12
Document File: 4 page(s) / 99K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed are a system and method for automatic lifeguard surveillance of a beach using crowd control technology coupled with swimmer identification and modeling for focused surveillance on poor swimmers.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 51% of the total text.

Page 01 of 4

System and Method for Automated Lifeguard

The problem of guarding crowded beaches with multiple swimmers is growing as urban populations in the world grow, and the public funding for adequate lifeguard coverage is limited.

A system and method are needed to improve surveillance of crowded swimming areas .

Disclosed are a system and method for automatic lifeguard surveillance of a beach using crowd control technology coupled with swimmer identification and modeling for focused surveillance on poor swimmers. The system provides swimmer risk classification from multinomial classification, as well as grouped swimmer and residual forecasting. The system also provides the average of Swimmer Info forecasting and grouped forecasting while adding swimming residual error. A forecasting sliding

window is developed based on swimmer risk.

The core components and process for the system and method include :

1. Segment video stream(s) from swimming area into swimmers, assigning swimmer identifiers (IDs)

2. Model the behavior over time of each swimmer, based on gesture, action recognition, and prior measures of correlates of these actions with swimming ability

3. Add external information, when available, to model swimmer (e.g., classes, years swimming, joint ailments, etc.)


4. Rank swimmers based on models according to ability


5. Differentially devote computational resources to swimmers for monitoring and

automatic lifeguard functions based on swimmers' ranked abilities

The system segments a video scene of a swimming area based on swimmers in the

water, models swimmers' actions based on several qualities, ranks identified swimmers by ability based on associated modeled qualities, and devotes computational resources to swimmers differentially based on each swimmer's rank (i.e. poor swimmers are closely watched). Time series forecasting looks several minutes into the future as well as a longer-term trend line to predict if a swimmer is in immediate danger or will be in the long term. External information such as classes taken, years swimming, frequency of swimming, joint ailments, etc. is input into a multinomial logistic regression model to classify a prior risk factor such as high risk, medium risk, and low risk. The combination of an ensemble of forecasters, Swimmer Info forecast, and a residual forecaster, produces an overall risk score. The higher the risk, the smaller the forecast window becomes for the individual until an alert is sent to a lifesaving group .

Figure 1: Automated Lifeguard architecture

1


Page 02 of 4

To classify a swimmer into high risk, medium risk, or low risk given prior predictors, a multinomial logistic regression model is built and applied. The k is the class label while the x is the feature vector for a swimmer. The beta values have been trained with training information.

Figure 2: Formula

The output of each k's class probability then becomes a predictor with real time features from a swimmer within a series of forecast...