Nothing Special   »   [go: up one dir, main page]
Scribd Logo
Upload

Welcome to Scribd!

  • 55 views

    Zachry Department of Civil and Environmental Engineering, Texas A&M University, College Station

    Uploaded by

    Asim Bashir Khajwal
    This document proposes a two-step methodology to enhance the reliability of crowdsourced damage assessments after disasters: 1) Microtasking to address unreliability from subjective tasks. 2) A maximum a-posteriori (MAP) model to address noise from participant responses. It applies this methodology in a case study of Hurricane Harvey, analyzing satellite imagery and photos tagged by 70 participants to infer building damage levels. The MAP model provides more accurate and ambiguous probabilistic damage estimates than majority voting, especially with unreliable participants. This improves the efficiency and reliability of preliminary post-disaster damage assessments.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Zachry Department of Civil and Environmental Engineering, Texas A&M University, College Station

    Uploaded by

    Asim Bashir Khajwal
    55 views1 page
    This document proposes a two-step methodology to enhance the reliability of crowdsourced damage assessments after disasters: 1) Microtasking to address unreliability from subjective tasks. 2) A maximum a-posteriori (MAP) model to address noise from participant responses. It applies this methodology in a case study of Hurricane Harvey, analyzing satellite imagery and photos tagged by 70 participants to infer building damage levels. The MAP model provides more accurate and ambiguous probabilistic damage estimates than majority voting, especially with unreliable participants. This improves the efficiency and reliability of preliminary post-disaster damage assessments.

    Original Description:

    Original Title

    Final Poster_CIR_AsimBKhajwal

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document proposes a two-step methodology to enhance the reliability of crowdsourced damage assessments after disasters: 1) Microtasking to address unreliability from subjective tasks. 2) A maximum a-posteriori (MAP) model to address noise from participant responses. It applies this methodology in a case study of Hurricane Harvey, analyzing satellite imagery and photos tagged by 70 participants to infer building damage levels. The MAP model provides more accurate and ambiguous probabilistic damage estimates than majority voting, especially with unreliable participants. This improves the efficiency and reliability of preliminary post-disaster damage assessments.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    55 views1 page

    Zachry Department of Civil and Environmental Engineering, Texas A&M University, College Station

    Uploaded by

    Asim Bashir Khajwal
    This document proposes a two-step methodology to enhance the reliability of crowdsourced damage assessments after disasters: 1) Microtasking to address unreliability from subjective tasks. 2) A maximum a-posteriori (MAP) model to address noise from participant responses. It applies this methodology in a case study of Hurricane Harvey, analyzing satellite imagery and photos tagged by 70 participants to infer building damage levels. The MAP model provides more accurate and ambiguous probabilistic damage estimates than majority voting, especially with unreliable participants. This improves the efficiency and reliability of preliminary post-disaster damage assessments.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 1

    Trustworthy Crowdsourcing for Rapid Disaster Damage Assessment: Addressing Uncertainty

    and Enhancing Reliability


    Asim Bashir Khajwal, Arash Noshadravan
    Zachry Department of Civil and Environmental Engineering, Texas A&M University, College Station
    INTRODUCTION CASE STUDY
    • Preliminary damage assessment (PDA) forms the basis for • The post-hurricane visual damage data from
    effective post-disaster planning, management, and fund- Hurricane Harvey is used.
    allocations.
    • Traditionally, PDA is carried out by expert inspection Advancing the effectiveness of crowdsourcing in • The data includes the satellite imagery, geo-
    tagged ground photographs.
    and reconnaissance teams.
    • Due to vast spatial extent, inaccessibility and limited post-disaster damage assessment by enhancing ƒ 60 buildings
    ƒ 70 participants
    time and resources, this process is very inefficient.
    Figure 1. Data sources available for PDA
    the cognition, content, and reliability of ƒ Expert-labels taken as the benchmark. Figure 5. Sample data compiled for 1 building.

    METHODOLOGY information gathered through public participation. RESULTS


    ¾ The probability distribution across different damage states is more ambiguous
    Two-fold procedure is proposed: in the case of the majority-vote heuristic as compared with our MAP model.
    • Micro tasking: To address the unreliability due to task subjectivity.
    • Maximum a-posteriori (MAP) model: To address the unreliability due to
    noisy participant responses. The development leads to improved:

    9 Efficiency
    in performing preliminary post-disaster damage assessment in a disaster-
    affected community.
    Figure 6. Probabilistic inference of the damage levels (a). MAP-based model (b). Majority-Vote.
    9 Reliability
    in leveraging citizen-driven assessments to inform the otherwise domain- ¾ The accuracy of the majority-vote based approach
    Figure 2. An illustration showing the methodology adopted in the present study.
    specific and subjective PDA task. declines with the increasing representation of
    unreliable participants.
    MATHEMETICAL SETUP

    Assumptions: 9 Flexibility
    ¾ The observed labels is the function of: in the integrating crowd-workers in larger frameworks for more advanced Figure 7. Effect of adversarial participants.
    9 Difficulty of the image - 1/ ∈ [0, ∞). automated damage assessment and decision support tools.
    9 Expertise of the labeler - ∈ (−∞, +∞).
    9 True label of the image -

    ¾ The labels given by labeler to image Figure 8. Variation of θ and ξ with number of adversarial participants
    are assumed to be generated as: Figure 9. Correlation between the mean values of
    1 parameters
    ( = )=
    1+
    CONCLUSION

    Scan to access the detailed paper ¾ The study improves the effectiveness of conducting and inferring citizen
    assessment of building damage states.
    asim.bashir@tamu.edu
    ¾ The model outperforms majority-vote heuristic and is effective in quantifying the
    probabilistic descriptions of overall damages.

    ACKNOWLEGEMENT
    Figure 3. Expectation Maximization (EM) algorithm This research is partially funded by Texas Sea Grant, Texas A&M University.
    Figure 4. Flowchart

    You might also like