Drone Dataset TH

ชุดข้อมูลตรวจจับอากาศยานไร้คนขับในบริบทไทย

A dataset programme for detecting and classifying drones under Thai environmental conditions — tropical haze, monsoon light, dense urban clutter and long-range sky backgrounds.

โครงการพัฒนา Dataset สำหรับการตรวจจับและจำแนก Drone ในบริบทประเทศไทย

View ResearchDemoArchitecturePaperDatasetCodeAPI

Problem

Public drone datasets are captured in temperate climates and short range, so models degrade badly on Thai sky, weather and background conditions.

Research Question

How much locally captured data is needed before detection performance stabilises across weather classes?

Objectives

Collect multi-environment, multi-weather drone imagery and video
Establish AI-assisted labeling with mandatory human verification
Publish versioned YOLO and COCO exports with full metadata

Methodology

01DATA COLLECTION
02DATA CLEANING
03AI AUTO LABELING
04HUMAN VERIFICATION
05DATASET VERSIONING
06MODEL TRAINING
07BENCHMARK
08DEPLOYMENT

Architecture

Capture RigsMulti-camera

Resolution and lens metadata recorded

IngestFFmpeg

Frame extraction and deduplication

Auto LabelingYOLO + SAM

Proposal generation only

VerificationHuman review

Every label confirmed by a person

VersioningDataset registry

Immutable releases

BenchmarkEval harness

Per-weather breakdown

Results & Benchmark

Weather classes covered: clear / haze / rain / dusk
Label policy: 100% human-verified

Limitations

· Coverage is currently biased toward central Thailand capture sites