Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Detect Body and Hand Pose with Vision
Search
Koki Hirokawa
July 01, 2020
Technology
970
0
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
Detect Body and Hand Pose with Vision
Koki Hirokawa
July 01, 2020
More Decks by Koki Hirokawa
See All by Koki Hirokawa
KMP の Swift export
kokihirokawa
0
840
SwiftUI移行のためのインプレッショントラッキング基盤の構築
kokihirokawa
0
430
App内課金におけるトラブルを劇的に減らすための取り組み
kokihirokawa
2
1.3k
ABEMAのリアーキテクチャ
kokihirokawa
5
5.5k
Other Decks in Technology
See All in Technology
Kiro入門|仕様駆動開発で変わるAI時代の開発スタイル
cmkudo
0
380
AI for Science時代を切り開く、政府の次世代HPC戦略の展望
gpuunite_official
0
250
ハッカソンで入賞した話 @ Findy LT
asari194617
0
1.5k
巨大気象データと戦う ― サロゲートモデル学習を高速化する圧縮技術
gpuunite_official
0
270
Eight Engineering Unit 紹介資料
sansan33
PRO
3
8.3k
Kiro WebとCloud Sessions
nagisa53
2
200
サイボウズ 開発本部採用ピッチ / Cybozu Engineer Recruit
cybozuinsideout
PRO
12
85k
GopherCon @シアトル に行ってきました
logica0419
0
210
Digitization部 紹介資料
sansan33
PRO
2
7.7k
LLM Internals: 언어 모델의 계보와 알고리즘 진화 (2023~2026)
inureyes
PRO
1
930
英語が話せなくてもKubeConスタッフに参加した話
yusuke427
2
1.1k
AIエージェントのためのデータ設計
daiz21
0
190
Featured
See All Featured
Neural Spatial Audio Processing for Sound Field Analysis and Control
skoyamalab
0
410
JAMstack: Web Apps at Ludicrous Speed - All Things Open 2022
reverentgeek
1
580
Highjacked: Video Game Concept Design
rkendrick25
PRO
1
440
Un-Boring Meetings
codingconduct
0
390
Avoiding the “Bad Training, Faster” Trap in the Age of AI
tmiket
0
210
From Legacy to Launchpad: Building Startup-Ready Communities
dugsong
0
310
Imperfection Machines: The Place of Print at Facebook
scottboms
270
14k
How to build a perfect <img>
jonoalderson
1
5.9k
Digital Projects Gone Horribly Wrong (And the UX Pros Who Still Save the Day) - Dean Schuster
uxyall
1
2.5k
Fashionably flexible responsive web design (full day workshop)
malarkey
408
67k
Taking LLMs out of the black box: A practical guide to human-in-the-loop distillation
inesmontani
PRO
3
2.4k
Responsive Adventures: Dirty Tricks From The Dark Corners of Front-End
smashingmag
254
22k
Transcript
Koki Hirokawa Detect Body and Hand Pose With Vision CA.swift
#12
ኍ߉ل / Koki Hirokawa / ͽʔͻΖ GitHub -> KokiHirokawa Twitter
-> @pihero13
ਓؒͷࢹ֮γεςϜʹ͓͍ͯߦ͑ΔλεΫΛίϯϐϡʔλͰ࣮ݱ͢Δ͜ͱΛٻ͢Δݚڀ Computer Vision ͱʁ
iOS 11 Ͱಋೖ͞ΕͨϑϨʔϜϫʔΫ إͷϥϯυϚʔΫݕग़ɾςΩετݕग़ɾόʔίʔυೝࣝɾը૾ೝࣝ ྨମݕग़ʹಠࣗͷ Core ML ϞσϧΛ༻͢Δ͜ͱՄೳ iOS /
Vision ͱʁ
ମͱखͷϙʔζΛݕͰ͖ΔΑ͏ʹͳͬͨ Detect Body and Hand Pose with Vision
͏ࣗࡱΓ͍Βͳ͍ ಛఆͷδΣενϟʔΛݕͯ͠γϟολʔΛΔ ׆༻ྫᶃ
δΣενϟʔͷྨ ֆจࣈΛΦʔόʔϨΠ ׆༻ྫᶄ
࣮ࡍͲͷ͘Β͍ͷύϑΥʔϚϯε͕ ग़ͤΔͷʁ
ͪ͜ΒαϯϓϧΞϓϦͱͯ͠͞Ε͍ͯ·͢ ۭதͰจࣈΛॻ͚ΔΞϓϦͰݕূ
ྲྀੴʹֆݫ͍͠Μ͡Όͳ͍ʁ
ԿΛඳ͍͍ͯΔ͔νϟοτͰͯͯΈͯԼ͍͞
ࢥͬͨΑΓઢ͕៉ྷʹॻ͚Δʂ ࢦΛ͍Α͘ಈ͔ͯ͠ݕͯ͘͠ΕΔʂ Beta ൛͚ͩͲɺ҆ఆ͍ͯ͠Δɻ ͨͩɺશͯͷࢦ͕ݟ͑ΔΑ͏ʹ͠ͳ͍ͱ্ख͘ݕͯ͘͠Εͳ͍ɻ ମݧͯ͠Έͨॴײ
ը໘ͷͰ্ख͘ػೳ͠ͳ͍͜ͱ͕͋Δ Χϝϥͷࢹํʹରͯ͠ฒߦͳखݕ͕͍͠ खାΛ͍ͯ͠Δ߹ݕ͕͍͠ Λखͱצҧ͍ͯ͠͠·͏͜ͱ͕͋Δ σόΠεݻఆ͞ΕΔඞཁ͕͋Δ ਫ਼ΛߴΊΔͨΊͷߟྀࣄ߲
࣮໘ͷʹҠΓ·͢
Request ͷ࡞ Handler ͷ࡞ & Handler ʹ Request Λ͢ ݁Ռͱͯ͠
Observation Λऔಘ͢Δ λεΫͷྲྀΕ
var handPoseRequest = VNDetectHumanHandPoseRequest() let handler = VNImageRequestHandler( cmSampleBuffer: sampleBuffer,
orientation: .up, options: [:] ) try! handler.perform([handPoseRequest]) let observation = handPoseRequest.results?.first as! VNRecognizedPointsObservation
var handPoseRequest = VNDetectHumanHandPoseRequest() let handler = VNImageRequestHandler( cmSampleBuffer: sampleBuffer,
orientation: .up, options: [:] ) try! handler.perform([handPoseRequest]) let observation = handPoseRequest.results?.first as! VNRecognizedPointsObservation
var handPoseRequest = VNDetectHumanHandPoseRequest() let handler = VNImageRequestHandler( cmSampleBuffer: sampleBuffer,
orientation: .up, options: [:] ) try! handler.perform([handPoseRequest]) let observation = handPoseRequest.results?.first as! VNRecognizedPointsObservation
var handPoseRequest = VNDetectHumanHandPoseRequest() let handler = VNImageRequestHandler( cmSampleBuffer: sampleBuffer,
orientation: .up, options: [:] ) try! handler.perform([handPoseRequest]) let observation = handPoseRequest.results?.first as! VNRecognizedPointsObservation
public func captureOutput( _ output: AVCaptureOutput, didOutput sampleBuffer: CMSampleBuffer, from
connection: AVCaptureConnection ) { var handPoseRequest = VNDetectHumanHandPoseRequest() let handler = VNImageRequestHandler(cmSampleBuffer: sampleBuffer, …) try handler.perform([handPoseRequest]) let observation = handPoseRequest.results?.first as! VNRecognizedPointsObservation } AVCaptureVideoDataOutputSampleBufferDelegate AVCaptureSession AVCaptureVideoDataOutput
֤ϥϯυϚʔΫͷऔಘ - VNRecognizedPointGroupKey Thumb Index Finger Middle Finger Ring Finger
Little Finger
֤ϥϯυϚʔΫͷऔಘ - VNRecognizedPoint
VNRecognizedPoint • ࠲ඪͷ৴པ • x ࠲ඪ • y ࠲ඪ •
ϥϕϧ
let thumbPoints = try observation.recognizedPoints(forGroupKey: .handLandmarkRegionKeyThumb) let thumbTipPoints = thumbPoints[.handLandmarkKeyThumbTIP]!
let x = thumbTipPoints.location.x let y = thumbTipPoints.location.y
let thumbPoints = try observation.recognizedPoints(forGroupKey: .handLandmarkRegionKeyThumb) let thumbTipPoints = thumbPoints[.handLandmarkKeyThumbTIP]!
let x = thumbTipPoints.location.x let y = thumbTipPoints.location.y
let thumbPoints = try observation.recognizedPoints(forGroupKey: .handLandmarkRegionKeyThumb) let thumbTipPoints = thumbPoints[.handLandmarkKeyThumbTIP]!
let x = thumbTipPoints.location.x let y = thumbTipPoints.location.y
None
VNDetectHumanHandPoseRequest ͷ maximumHandCount Λௐ͢Δ͜ͱͰෳͷख Λݕ͢Δ͜ͱ͕Մೳ σϑΥϧτ 2 ύϥϝʔλͷΛେ͖͘͢Εͦͷ͚ͩϨΠςϯγʹӨڹ͕ੜ͡Δ ͘ઃఆ͢ΕɺͦͷҎ্ͷखݕग़͞Εͳ͘ͳΓύϑΥʔϚϯε্͕Δ ϨΠςϯγʹؔ͢Δҙࣄ߲
BODY POSE ΄΅ಉ͡ॲཧʹͳΓ·͢
ARKit ͷϞʔγϣϯΩϟϓνϟʔͱͷҧ͍
͝ਗ਼ௌ͋Γ͕ͱ͏͍͟͝·ͨ͠
ग़య ʮDetect Body and Hand Pose with Visionʯ ؔ࿈ηογϣϯ ʮCreate
ML Λ༻ͨ͠ΞΫγϣϯྨࢠͷ࡞ʯ ʮComputer Vision API ͷ୳ٻʯ