Idea #3

EEG-Augmented Witness Triage for Suspect Search (RSVP + CCTV)

Brief:

A two-stage system to help investigators prioritise potential suspects from large CCTV pools using “brain-in-the-loop” ranking.

Computer Vision stage: detect + track faces in CCTV, cluster identities, and generate a candidate set of representative face frames.
Witness RSVP + EEG stage: witnesses view rapid sequences of candidate faces/frames (RSVP/oddball paradigm). An EEG classifier detects recognition-like ERP responses (e.g., P300-family effects) and re-ranks candidates for manual review.

The output is a shortlist for investigators, not a definitive identification.

Aim:

Reduce time-to-review and missed leads in CCTV-heavy investigations by using EEG as an implicit feedback channel to boost candidates that match a witness’s memory.

Deliverables:

CCTV face pipeline: detection -> tracking -> clustering -> representative frame selection
RSVP witness UI with controlled stimulus design (foils, catch trials, breaks)
Real-time EEG pipeline (stream -> preprocess -> features -> classifier -> ranking)
Evaluation on controlled “mock witness” datasets: speed/recall vs manual review, plus workload/UX metrics

Novelty:

Brings together modern CV candidate generation with ERP-based implicit recognition to create a practical, investigatory triage tool (lead prioritisation), rather than a courtroom "truth detector".

Research Questions:

Main:

Does EEG-augmented witness triage improve suspect-candidate prioritisation (time-to-find, recall/false positives) compared to standard witness review alone?

Mechanism:

Can recognition-like ERP responses (P300-family) be detected reliably enough in RSVP for ranking, and how sensitive is this to familiarity/contamination and individual differences?

UX:

Is the RSVP + EEG workflow acceptable for witnesses (fatigue, stress, perceived intrusiveness), and does it improve confidence/clarity compared to traditional lineups?

Scope / Assumptions:

Output is investigative prioritisation only; no claims of guilt or proof.
EEG is treated as noisy; system falls back to behaviour-only ranking when signal quality is poor.
Strong controls are required to reduce false positives from “mere familiarity” and media contamination (foils/catch trials and careful stimulus selection).
Initial prototype uses controlled datasets and mock-witness protocols before any real-world deployment discussion.

References:

Brain activity-based image classification from rapid serial visual presentation (PubMed)
- Journal article (peer-reviewed)
- Published Oct 2008 - Nima Bigdely-Shamlo, Andrey Vankov, Rey R. Ramirez, Scott Makeig (IEEE Transactions on Neural Systems and Rehabilitation Engineering) (PubMed)
- https://pubmed.ncbi.nlm.nih.gov/18990647/ (PubMed)
An Iterative Framework for EEG-based Image Search: Robust Retrieval with Weak Classifiers (PLOS)
- Journal article (peer-reviewed)
- Published Aug 20, 2013 - M. Ušćumlić, R. Chavarriaga, J. del R. Millán (PLOS ONE) (PLOS)
- https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0072018 (PLOS)
EEG Dataset for RSVP and P300 Speller Brain-Computer Interfaces (Nature)
- Data descriptor / dataset paper (peer-reviewed; open access)
- Published 08 July 2022 - Kyungho Won, Moonyoung Kwon, Minkyu Ahn, Sung Chan Jun (Scientific Data) (Nature)
- https://www.nature.com/articles/s41597-022-01509-w (Nature)
Method for enhancing single-trial P300 detection by introducing the complexity degree of image information in rapid serial visual presentation tasks (PLOS)
- Journal article (peer-reviewed; open access)
- Published Dec 28, 2017 - Zhimin Lin, Ying Zeng, Li Tong, Hangming Zhang, Chi Zhang, Bin Yan (PLOS ONE) (PLOS)
- https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0184713 (PLOS)
The effect of target and non-target similarity on neural classification performance: a boost from confidence (PubMed)
- Journal article (peer-reviewed)
- Published Aug 5, 2015 - Amar R. Marathe et al. (Frontiers in Neuroscience) (PubMed)
- https://pubmed.ncbi.nlm.nih.gov/26347597/ (PubMed)