|
Adaptive, Automated Real-time Event/Target Detection using Supervised Learning
Navy SBIR 2012.1 - Topic N121-017 NAVAIR - Ms. Donna Moore - [email protected] Opens: December 12, 2011 - Closes: January 11, 2012 N121-017 TITLE: Adaptive, Automated Real-time Event/Target Detection using Supervised Learning TECHNOLOGY AREAS: Information Systems, Battlespace ACQUISITION PROGRAM: PMA 281 RESTRICTION ON PERFORMANCE BY FOREIGN CITIZENS (i.e., those holding non-U.S. Passports): This topic is "ITAR Restricted". The information and materials provided pursuant to or resulting from this topic are restricted under the International Traffic in Arms Regulations (ITAR), 22 CFR Parts 120 - 130, which control the export of defense-related material and services, including the export of sensitive technical data. Foreign Citizens may perform work under an award resulting from this topic only if they hold the "Permanent Resident Card", or are designated as "Protected Individuals" as defined by 8 U.S.C. 1324b(a)(3). If a proposal for this topic contains participation by a foreign citizen who is not in one of the above two categories, the proposal will be rejected. OBJECTIVE: Develop an automatic, adaptive event detection tool that allows supervised learning to minimize operator involvement and achieve near real-time performance. DESCRIPTION: Current Digital Camera Recording System (DCRS) and Unmanned Air Vehicle (UAV) control systems ingest and display imagery data from various sources that include 8mm tape, other magnetic media, solid state devices, and Full Motion Video (FMV) via data link. Some of the events and targets are identified during the initial image acquisition, but when transferring, analysts must watch the transfer to the DCRS to mark the events on the DCRS, which is a time consuming activity. UAV operators often watch live imagery feeds from a data link, a time and bandwidth consuming activity which can result in large inefficiencies and missed events. Previous efforts focused on Automatic Target Recognition (ATR) to identify specific entities. Other efforts used basic change detection algorithms over long time periods to identify events. ATR efforts along these lines have produced some successes, but have not captured a capability useful across multiple environments with varying situations. Proposers are encouraged to review unclassified technical reports highlighting development and progress which can be gleaned from Proceedings of SPIE (International Society for Optics and Photonics) conferences, symposia reports and other relevant IEEE (Institute of Electrical and Electronics Engineers) conference proceedings. (see References) For this topic we define an event as the interaction of one or more objects in an environment (e.g. a small truck departing a parking lot, a person dismounting a vehicle). Each object has basic properties such as size, color, speed, location, and direction of movement. Events can also have properties (e.g. action, time, and location). In order to minimize the topic complexity, this effort will be constrained to identifying the activity of the object in relation to other objects and their locations. Object recognition through formal ATR processes is not expected. The development of this effort will focus on a specific high priority event set identified by current sensor operators. To constrain the overall research and development, the effort will be limited initially to an event of critical importance (e.g. person exiting vehicle), with specific object and event properties. The goal of the effort is to recognize and alert the user of 90 percent of the instances of a specified event. These detections should be encoded as imagery and video metadata. Based on these technical advancements this topic concentrates and will direct the innovative development of a novel software tool capable of performing automatic event recognitions as outlined above. The tool should have the capability to accept instruction (i.e. real-time supervised learning) regarding automated recognition of these particular and/or similar events in future occurrences. This software tool should be trainable with an a priori defined event database similar to the biological process or other advanced algorithmic techniques (as defined by proposer(s)) as a baseline for automatic real-time detection. In future phases, the software tool should allow real time sharing of the event database between analyst stations. A Service Oriented Architecture software environment is preferred. PHASE I: Design and prove the feasibility of a software tool that is not only able to automatically detect time critical events but must be �trainable� by analysts so as to add previously unidentified ad hoc events when indicated. Detections should be encoded as imagery and video metadata and highlighted to users. PHASE II: Develop and demonstrate a prototype based upon the design and innovations developed in Phase I. PHASE III: Transition the tool developed to the DCRS and/or Unmanned Vehicle Common Control Station and adapt to and implement in, commercial applications. PRIVATE SECTOR COMMERCIAL POTENTIAL/DUAL-USE APPLICATIONS: The system developed could readily be transitioned to many persistent surveillance applications such as border, coastal and harbor security monitoring. Additionally, private sector security monitoring systems can greatly benefit from this technology. REFERENCES: 2. Dikmen, M., Ning, H., Lin, D., Cao, L., Le, V., Tsai, S., Lin, K., Li, Z., Yang, J., Huang, T. (2009). Surveillance Event Detection. http://www-nlpir.nist.gov/projects/tvpubs/tv8.papers/ifp-uiuc-nec.pdf 3. Seo, H.J. & Milanfar, P. (2010). Visual Saliency for Automatic Target Detection, Boundary Detection and Image Quality Assessment. http://users.soe.ucsc.edu/~milanfar/publications/conf/Saliency_ICASSP_Final.pdf KEYWORDS: Automatic Target Recognition; supervised learning; service oriented architecture; software; UAV control station; Autonomous event detection
|