Crowd Detection and Accessibility Navigation

Real-time crowd & proximity alerts for mobility-impaired travellers.

YOLOv8 person detection + bounding-box proximity heuristics, served as React → Spring Boot → FastAPI/YOLO. Fine-tuned on JRDB (Stanford JackRabbot data, Monash ERC host).

Quick start (Docker)

cd application
mkdir -p models && cp inference-service/best.pt models/best.pt   # Windows: copy inference-service\best.pt models\best.pt
docker compose up --build

Open http://localhost → Start Detection → allow webcam. Needs application/models/best.pt (shipped copy in inference-service/).

Service	URL
Frontend	http://localhost
API	http://localhost:8080/api/v1/analyze-frame
Inference (internal)	http://localhost:9000/health

Local dev: run inference-service, backend, and frontend on ports 9000 / 8080 / 5173.

Repository

train/          YOLO training, preprocessing, notebooks
application/    React UI, Spring API, FastAPI inference (docker-compose.yml)
infra/          SageMaker launcher, Docker wrapper
docs/           Specs, architecture, runbooks, diagrams
data/           Dataset root (DVC) — see docs/DATA.md

Python setup (repo root): python -m venv .venv → activate → pip install -r requirements.txt
Train: cd train && python scripts/train_yolo.py (default yolov8m.pt, 8:1:1 split)
Data: dvc pull — details in docs/DATA.md
Cloud train: infra/sagemaker/sagemaker_launch.py (default ml.g5.xlarge)

Documentation

Doc	Topic
docs/REQUIREMENTS.md	FR / NFR
docs/API_SPEC.md	REST + inference API
docs/TechSpec.md	Stack & deployment
docs/PRD.md	Product vision
docs/DATA.md	DVC & dataset layout
docs/reports/Final_Training_Report.md	Training results
docs/RELEASE_v2.6.0.md	v2.6.0 release notes
docs/RELEASE_v2.5.1.md	v2.5.1 release notes
docs/RELEASE_v2.5.0.md	v2.5.0 release notes
docs/architecture/ · docs/diagrams/ · docs/decisions/	Architecture & ADRs

Team

Name	ID	Focus	Commits*	Share*
Jungwook Van	25167747	YOLO training (lead), inference & alerting	193	97.0%
Phoi Gia Vuong	25736012	Data pipeline, preprocessing, documentation	3	1.5%
Chihyun An	14707133	Frontend UI	3	1.5%

*From git log --all --no-merges on main (excludes merge commits and bots). Salieri009 + Jungwook Van are the same contributor (kordalek@* emails). Last updated: 2026-06-18.

Assignment (42028 S1 2026)

• 3 students per group; model training required (transfer learning OK).
• Parts A & E & G submitted individually; B–D one per team. Oral defense (G) is mandatory.
• Filename: Assignment-3-<Part>-<Name>-<ID>.pdf

Questions → subject coordinator by email.

UTS Deep Learning (42028) · Semester 1, 2026