0325-06_-_York_University_Services_-_SOWv3.pdf

PROTOTYPE RECOGNITION OF THREATS IN AN AUTONOMOUS PAINTBALL

PLAYER SCENARIO STATEMENT OF WORK (SOW)

1.0 BACKGROUND

This Statement of Work describes the tasks and responsibilities to be fulfilled by York University., (hereafter referred to as the contractor), to provide C++ code, technical report, and a technical brief to the Naval Surface Warfare Center, Dahlgren

Division (NSWCDD) to assist detecting relevant target entities from visual imagery.

1.1 Scope

The project will examine the problem of threat identification in complex images for the goal of providing target information to an autonomous weapons mount surrogate. This will involve analysis of saliency characteristics of threats and threat distractors (including camouflage and other image components that obscure figure-ground segregation processes). The contractor will develop a prototype system that will segregate figure (potential threat) from ground (non-threatening image background) and perform a simple classification into a small number of threat types. Threat types will be ranked based on confidence of the classification scheme. Centroid location estimates will be included. The goal of this prototype system is to demonstrate the potential and efficacy of the overall strategy as a precursor for a more sophisticated and robust future effort.

3.0 REQUIREMENTS

3.1 Attentional Models

The contractor shall examine each of these to see which are relevant for the task and specify prototypical algorithms. The use of Saliency methods for determining region-of-interest, features-of-interest and time-of-interest play will be leveraged. The findings shall be documented in a technical report (CDRL A002).

3.2 Classification Models

Current models for object classification and categorization exhibit excellent performance results. Region-of-interest determinations via saliency maps are not the only strategy and overall attention needs to be completely integrated into the overall processing stream. Models such as Convolutional Nets or

Deep Learning Networks are strictly feed-forward models and make a variety of assumptions that prevent attentive processing. The contractor shall examine such models within the context of target detection and determine which will be the best for redefinition into an attentive classification process. The contractor shall develop a prototype of this new model and provide an initial proof-of-concept evaluation (CDRL A002).

3.3 Prototype

The contractor shall design, implement and test a prototype system that will accept image sequences and yield classification and localization results for the scene elements present. This proof-of-concept prototype shall demonstrate application and implementation of the developed attention and classification models, allowing them to be compared to previous models in terms of classification and localization performance (CDRL A003).

3.4 Bi-weekly IPT Meetings

The contractor shall participate and report on status in progress meetings with the SAF-T Perception Integrated Product

Team (IPT). The contractor shall participate in these meetings via teleconferences.

3.5 Program Review Meeting

The Contractor shall provide a brief (CDRL A001) to the SAF-T project team at the Office of Naval Research (ONR) Supervised

Autonomous Fires Technology (SAF-T) Program Review. The meeting will be held at NSWCDD located in Dahlgren, Virginia. The contractors shall be authorized to travel to this event which is scheduled for up to 2 days.

4.0 GOVERNMENT FURNISHED INFORMATION (GFI)

4.1 Any analysis of updates to the SAF-T architecture by the

Government will be provided to the Contractor on an as needed basis.

4.2 Office of Naval Research (ONR) Code 30 SAF-T final Program

Review dates will be provided to contractor as soon as verified by the Government but not later than (NLT) three (3) weeks prior to the review.

5.0 DELIVERABLES

5.1 Technical Brief

The Contractor shall provide a technical brief to support the

ONR Code 30 SAF-T Program Review tentatively scheduled for 23

September 2014 (CDRL A001).

5.2 Technical Report

The Contractor shall deliver a technical report (CDRL A002).

5.3 Technical Data Package

The Contractor shall provide C++ source code and working copy of the software (CDRL A003).

6.0 TRAVEL

The contractor will be required to travel in performance of this contract. The number of trips and types of personnel traveling shall be limited to the minimum required to accomplish work requirements and shall be coordinated with the technical point of contact. All travel shall be conducted in accordance with the FAR 31.205-46, “Travel Costs”, and the Federal Travel

Regulations (FTR). Travel by two (2) personnel to NSWCDD located in Dahlgren, Virginia and University of Manitoba located in Winnipeg, Manitoba Canada is anticipated for two days each trip.

7.0 TECHNICAL POINT OF CONTACT

To be determined at award.