Browse Prior Art Database

PREDICTING MOVEMENT OF MULTIPLE, DISTINGUISHABLE TARGETS USING SPARSELY DISTRIBUTED SENSORS

IP.com Disclosure Number: IPCOM000245729D
Publication Date: 2016-Apr-01
Document File: 6 page(s) / 506K

Publishing Venue

The IP.com Prior Art Database

Related People

Giorgi Guliashvili: AUTHOR [+4]

Abstract

A methodology is presented that can predict future displacement of targets, using a very small amount of training data, when the targets exhibit a certain kind of behavior, particularly when they have certain destination points inside a tracking area, which might change gradually, and after reaching the destinations the targets leave a tracking area.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 33% of the total text.

Predicting movement of multiple, distinguishable targets using sparsely distributed sensors

AUTHORS: 

Giorgi Guliashvili

Kristen Wright

Hugo Latapie

Rodolfo Milito

CISCO SYSTEMS, INC.

ABSTRACT

A methodology is presented that can predict future displacement of targets, using a very small amount of training data, when the targets exhibit a certain kind of behavior, particularly when they have certain destination points inside a tracking area, which might change gradually, and after reaching the destinations the targets leave a tracking area.

DETAILED DESCRIPTION

             Predicting displacement (direction) of multiple, distinguishable targets, using sparsely distributed binary sensors in an unfriendly environment, is challenging.  For example, in a protected animal park, guards of the park are charged with stopping poaching of animals.  A system is needed to assist guards in detecting poachers using artificial intelligence generated displacement prediction to find poachers. Reliable prediction is important because poachers will not go near guard stations. Guards will need a reasonable time in order to go to the location where a poacher was detected.

There are several restrictions, which make this difficult, and necessitate a creative approach.

1)                 The park area is hundreds of square miles. 

2)                 There is only local Internet in the area.

3)                 Sensors need to be battery-powered, so that advanced, energy-consuming, high Internet traffic generator devices cannot be used.

4)                 Due to the nature of the problem, sample data (number of poachers) is small and is increasing slowly.

As every undetected poaching event is very expensive damage, the model used by the system needs to use training data efficiently and generate high quality predictions from the outset. Generally, most of the machine learning predictive models are very complex and needs a very large amount of training data. But in that case, such data simply does not exist. This means a trade-off needs to be done between model complexity and prediction quality. If the model were very complex, it would over-fit on several samples and so would need a large amount of data (which does not exist).

This problem is solved with one very important exploited fact about poachers, without conceding much quality of prediction. This solution can be applied in many other technical areas.

Exploited/used fact is that, there are some critical locations and the targets are going to those locations semi-optimally, and then moving out of the covered area. In the case of poaching, a critical location is where animals are mostly located. In the case of tracking humans in a museum, a critical location would be near a very popular painting, for example.  In the case of general human tracking, a critical location would be a working location, home area, such as bathrooms, etc. Exploiting this fact will not overly restrict or limit the solution. 

A model abstracts everything as a target consisting of an obje...