Automated Streams Analysis for Public Safety (ASAPS) Prize Challenge: Contest 1
Develop tools and capabilities to detect and analyze emergency events from live streaming multimodal public safety data
Department of Commerce - National Institute of Standards and Technology
Type of Challenge: Software and apps, Ideas, Analytics, visualizations, algorithms, Scientific
Submission Start: 04/13/2020 09:00 PM ET
Submission End: 08/07/2020 05:00 PM ET
The Public Safety Communications Research Division of the National Institute of Standards and Technology is conducting the Automated Streams Analysis for Public Safety (ASAPS) Prize Challenge, authorized under Section 105 of the America COMPETES Reauthorization Act of 2010 (Pub. L. No 111-358), as codified in 15 U.S.C. § 3719.
This challenge anticipates awarding prizes to challenge contestants who submit innovative approaches that address information extraction across ASAPS multimodal data streams and perform fusion in real time across extracted data to generate actionable data reports about simulated and staged emergency events. One of the most important objectives of this program is to demonstrate how data can be used to enable decision support tools and capabilities for public safety. This challenge is a nationwide innovation contest taking place throughout 2020, 2021, and 2022. The challenge is anticipated to be administered in four discrete contests. In the context of multiple contests, these are the official rules for Contest 1 only.
Contest 1 will focus on developing innovative “plans of attack” focused on data extraction and fusion, refining and focusing the details relevant to the ASAPS Challenge’s problem statement, partnering between contestants to form highly qualified teams, and encouraging collaborative and creative thought around potential approaches to developing solutions. NIST anticipates awarding up to $150,000 in cash to winning contestants in Contest 1. In addition, contestants will be familiarized with the development data set and version 1.0 of the test harness and encouraged to provide feedback for potential updates in advance of the second contest.
Pre-registration begins: February 10, 2020
Open to submissions: April 13, 2020
Submission deadline: August 7, 2020 @ 5pm ET
Reviewing & Judging: August 10 to September 14, 2020
Winners Announced: September 17, 2020
The area of real-time automated analytics is understudied – especially with regards to the analytic fusion of multiple sources of unstructured information in real-time. The future of public safety demands automated solutions to analyze live data, provide improved situational awareness, and alert and inform decision making.
Extracting information from video, audio and textual communications, social media, and sensors presents a significant analytic challenge. It is virtually impossible for humans to co-analyze these data streams in real-time, and, currently, there are few automated tools to analyze these types of unstructured data in real-time and virtually no automated tools to create dynamic analytic models of public safety emergencies as emergency situations evolve.
With the maturation of artificial intelligence (AI) technologies, specialized architectures and processors for AI applications, cloud computing, high-speed communications networks and streaming data resources and protocols, it is now possible to create automated tools to assist public safety organizations in analyzing the increasingly large amounts of information at their fingertips to help them make the most informed decisions based on a comprehensive analysis of the data available. Such tools will help distill, structure, and present the quickly evolving information from a multitude of data streams to public safety organizations in a way that supports time-critical decision making and serves as an essential force multiplier.
Due to the multitude of sensors and devices being used in public safety along with communications data collected in operations and contributed by the public, and the multi-modal nature of all of these data streams, transforming data, especially unstructured data, into information in real-time is one of the key technology gaps that must be bridged to achieve advances in this area. The available technology is insufficient to bridge this technological gap because:
Most automated data analysis tools have been developed for use with archival data and focus on post-event forensic analysis.
Few analytic technologies focus on a moving representation of live streaming data that is updated over time, and that can be used to show temporal changes in the analysis.
Tools that make data available to first responders in real-time are heavily weighted towards a visual representation of responder location and structured reporting of known emergency communications.
Few public safety emergency analysis tools exist that provide an effective real-time automated analysis and fusion of information from unstructured public safety communications streams and sensor- and citizen-generated data streams.
Note that while public safety organizations must often grapple with answering the who, what, when, and where of emergencies, the research supported by this Challenge focuses only on the what, when, and where aspects of streaming data analysis and specifically excludes identity-related technologies such as facial and voice biometrics which are considered out of scope for this research program - with the exception of technology to maintain scene-relative identification of people/vehicle/object/place entities involved in emergencies in order to associate activities with the entities they are involved with. This focus of the Challenge is because the area of identifying the situational and time-critical evolving characteristics of an emergency across information streams is under-researched. This Challenge seeks to focus research on analyzing a plethora of fast-moving data streams for life/infrastructure-critical situational information to support both effective and efficient response.
Public safety organizations are inundated with live streaming data from a variety of sources. This live streaming data generally falls into two categories - some of this data is structured, while much of it is unstructured. Structured data that resides in a fixed field within a record or file (the kind typically found in spreadsheets or relational databases) is formatted in a way so that it can be readily queried. By contrast, unstructured data (e.g., video and audio) generally does not have a pre-defined data model or is not organized in a pre-defined manner and, therefore, is not readily searchable. The lack of a defined data model prevents the data from being automatically interpreted and used by automated decision support applications without significant human intervention.
In this first contest, NIST PSCR is inviting a novel multidisciplinary community to share its vision for how to analyze numerous live data streams, understand the data, and transform it into information to support public safety decisions. We are looking for ideas that address the following analytic tasks as well as an overall extraction to analysis systems platform approach to the contest elements. The core analytic tasks for this contest can be broadly described as:
Information extraction across data streams and data sources: Structured information is continuously extracted from each of the data streams that are necessary to support downstream analysis and detection of emergency events.
Information fusion across extracted data and generation of live/dynamic information representation: Joint inferencing is performed across all of the information in the extracted streams to create an event-level analysis of an emergency and store that information in a dynamically evolving structured data representation that can be utilized at any time during the life of an emergency.
Automated emergency event analysis:Highly actionable analytic reports about emergency events are generated from the extracted/fused structured information (as described in the above topics) into a form that can be readily rendered into actionable user-readable reports in real-time and in both ongoing and summary forms.
Success in this ideation contest is dependent upon a clear vision of the components, platform, and novel approaches that fulfill the functions of real-time information extraction, live dynamic information fusion and representation, and evolving emergency event analysis.
Figure 1. A rendering of the flow of streaming data through a conceptual ASAPS system. Contest 1 focuses ONLY on the extraction and core analysis components of the system. Later contests will expand the focus to include the end user interfaces to the analytic information.
This figure is a simplified depiction of the flow of information through an analytic process and not prescriptive of the approaches that might be developed to address the challenge.
To provide some additional clarity on the context and make-up of the data, see tables for examples of public safety data types and emergency events.
A total prize purse of up to $150,000 will be awarded to the winners, as follows:
Topic 1: Information extraction across ASAPS streams and data sources across 1 or more data categories: $30,000
Topic 2: Information fusion across extracted data and generation of live/dynamic information representation, across 3 or more data categories: $30,000
Topic 3: Automated emergency event analysis across 3 or more data stream categories, and preferably across all data categories: $30,000
Topic 4: Extraction-to-analysis systems approach across all data categories: 2 X $30,000
View the Official Rules document for detailed rules on all aspects of this challenge.
Rules regarding submissions, as well as additional terms and conditions, begin on Page 6 of the document.
Submissions will initially be screened for completeness and compliance with the objectives and Official Rules of this challenge. A submission that fails to meet the compliance requirements listed below will be disqualified and will be ineligible to compete in this challenge. Submissions that pass the initial compliance review will be evaluated by a panel of subject matter experts and scored by a panel of judges. An evaluation of a submission by the Judging panel does not constitute NIST’s final determination of the contestant or submission eligibility.
To pass initial screening, submissions must:
Discuss the current technology landscape
Identify the technology gaps
Describe the proposed approach to bridging the identified gaps
Present the expertise and experience represented by the team and discuss how it aligns with the proposed approach
Be written in English
Submissions will be evaluated based on the criteria in this table. Scores will not be provided to the contestants. In the case of a tie, the Judging panel will decide on the winners.
In addition to the criteria listed in the table, submissions will also be evaluated to ensure they meet the requirements for the number of data categories (see Table 1 for data stream categories) required for each topic area. The requirements are as follows:
Topic 1: Information Extraction across ASAPS streams and data sources Submitted ideas must address information extraction across one or more data stream categories.
Topic 2: Information fusion across extracted data and generation of live/dynamic information representation Submitted ideas must address information fusion and information representation from 3 or more data stream categories.
Topic 3: Automated emergency event analysis Submitted ideas must address emergency event analysis across 3 or more data stream categories. Preference will be given to ideas that span all data stream categories.
Topic 4: Extraction-to-analysis systems approach Submitted ideas must address an extraction-to-analysis systems approach that uses all data stream categories.
How To Enter
Register as a contestant on HeroX.com and complete your submission using the online portal by the Submission Deadline.
Each individual, team, or organization is allowed only one entry.
There are four topics addressed in this contest and you may respond to one or more of these topics within your submission form.