Antigravity Agent
f8b21af4be
10_Wiki/Topics 대규모 정리: - 오류 캡처/미완성 stub 문서 227개 제거 - 교차폴더 중복 43클러스터 병합 (63파일 → redirect) - 링크명 정규화: 깨진 링크 수정·redirect 직결·개념 매핑 ~2,400건 - 카테고리 MOC 6개 신규 생성 - Graph 섹션 미해결 related-keyword 링크 10,058건 제거 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
133 lines
4.4 KiB
Markdown
133 lines
4.4 KiB
Markdown
---
|
|
id: wiki-2026-0508-깊이-지각-depth-perception
|
|
title: 깊이 지각(Depth perception)
|
|
category: 10_Wiki/Topics
|
|
status: verified
|
|
canonical_id: self
|
|
aliases: [Depth Perception, Stereopsis, 입체시]
|
|
duplicate_of: none
|
|
source_trust_level: A
|
|
confidence_score: 0.9
|
|
verification_status: applied
|
|
tags: [perception, vision, vr, graphics, psychology]
|
|
raw_sources: []
|
|
last_reinforced: 2026-05-10
|
|
github_commit: pending
|
|
tech_stack:
|
|
language: GLSL / C++
|
|
framework: OpenGL / Vulkan / Unity
|
|
---
|
|
|
|
# 깊이 지각(Depth perception)
|
|
|
|
## 매 한 줄
|
|
> **"매 brain 의 multiple cues 의 fusion 의 3D world reconstruction."**. 매 binocular (stereopsis) + monocular (parallax, perspective, occlusion) cues 의 combination, 매 VR/AR 의 design 의 foundation, 매 stereo rendering / depth map 의 graphics 의 핵심.
|
|
|
|
## 매 핵심
|
|
|
|
### 매 binocular cues
|
|
- Stereopsis: 매 두 eye 의 disparity (~63mm IPD).
|
|
- Convergence: 매 eye muscles 의 toe-in angle.
|
|
- Effective range: 매 ~10m 까지.
|
|
|
|
### 매 monocular cues
|
|
- Motion parallax: 매 가까운 물체 의 빠른 이동.
|
|
- Linear perspective: 매 parallel lines 의 vanishing point.
|
|
- Occlusion: 매 앞 의 물체 의 뒤 의 물체 가림.
|
|
- Relative size / texture gradient / aerial perspective.
|
|
- Accommodation: 매 lens focus 의 muscle feedback.
|
|
|
|
### 매 응용
|
|
1. Stereoscopic 3D rendering (VR, 3D cinema).
|
|
2. Depth-based effects (DOF, fog, SSAO).
|
|
3. Computer vision depth estimation (MiDaS, Depth Anything v2).
|
|
4. Photogrammetry / 3D reconstruction.
|
|
|
|
## 💻 패턴
|
|
|
|
### Pattern 1 — Stereo camera setup (Unity)
|
|
```csharp
|
|
public class StereoCamera : MonoBehaviour {
|
|
public float ipd = 0.063f; // 63mm
|
|
public Camera leftCam, rightCam;
|
|
void LateUpdate() {
|
|
leftCam.transform.localPosition = new Vector3(-ipd/2, 0, 0);
|
|
rightCam.transform.localPosition = new Vector3( ipd/2, 0, 0);
|
|
}
|
|
}
|
|
```
|
|
|
|
### Pattern 2 — Anaglyph 3D shader (GLSL)
|
|
```glsl
|
|
// Red-cyan anaglyph
|
|
vec3 left = texture(leftEye, uv).rgb;
|
|
vec3 right = texture(rightEye, uv).rgb;
|
|
fragColor = vec4(left.r, right.g, right.b, 1.0);
|
|
```
|
|
|
|
### Pattern 3 — Depth-from-stereo (block matching, OpenCV)
|
|
```cpp
|
|
auto stereo = cv::StereoBM::create(16, 15);
|
|
cv::Mat disparity;
|
|
stereo->compute(leftGray, rightGray, disparity);
|
|
// depth = baseline * focal / disparity
|
|
```
|
|
|
|
### Pattern 4 — Monocular depth estimation (Depth Anything v2)
|
|
```python
|
|
from transformers import pipeline
|
|
depth = pipeline("depth-estimation", model="depth-anything/Depth-Anything-V2-Large")
|
|
result = depth(image) # returns depth map
|
|
```
|
|
|
|
### Pattern 5 — Vergence-accommodation conflict mitigation (varifocal HMD)
|
|
```cpp
|
|
// Track gaze depth, drive lens focal distance
|
|
float gazeDepth = eyeTracker.getVergenceDistance();
|
|
varifocalLens.setFocalDistance(gazeDepth);
|
|
```
|
|
|
|
### Pattern 6 — DOF (depth of field) post-process
|
|
```glsl
|
|
float depth = texture(depthTex, uv).r;
|
|
float blur = abs(depth - focusDepth) * dofStrength;
|
|
fragColor = mix(sharpColor, blurredColor, clamp(blur, 0, 1));
|
|
```
|
|
|
|
## 매 결정 기준
|
|
| 상황 | Approach |
|
|
|---|---|
|
|
| 매 VR rendering | Stereo cameras + IPD calibration |
|
|
| 매 single image depth | Depth Anything v2 / MiDaS |
|
|
| 매 dual camera depth | Stereo block matching / SGBM |
|
|
| 매 LiDAR available | Direct depth (no estimation) |
|
|
| 매 NeRF / Gaussian Splatting | Multi-view consistency loss |
|
|
|
|
**기본값**: 매 VR 의 stereoscopic + 매 single image 의 Depth Anything v2.
|
|
|
|
## 🔗 Graph
|
|
- 부모: [[Computer Vision|Computer-Vision]]
|
|
- 변형: [[Stereopsis]]
|
|
- 응용: [[가상현실(VR)]]
|
|
- Adjacent: [[Vergence-Accommodation-Conflict]]
|
|
|
|
## 🤖 LLM 활용
|
|
**언제**: 매 VR rendering pipeline 의 design, 매 depth cue 의 trade-off 분석, 매 monocular depth model 의 selection.
|
|
**언제 X**: 매 individual user 의 stereo blindness 의 diagnosis (medical), 매 hardware-specific IPD calibration.
|
|
|
|
## ❌ 안티패턴
|
|
- **Fixed IPD**: 매 user 마다 의 IPD 의 variation (55-72mm) 무시 → eye strain.
|
|
- **Excessive parallax**: 매 screen edge 의 reverse stereo → headache.
|
|
- **VAC (Vergence-Accommodation Conflict)**: 매 fixed-focal HMD 의 close objects 의 discomfort.
|
|
- **Monocular cue 만 의존**: 매 depth ambiguity 의 cause.
|
|
|
|
## 🧪 검증 / 중복
|
|
- Verified (Goldstein "Sensation and Perception", SIGGRAPH papers).
|
|
- 신뢰도 A.
|
|
|
|
## 🕓 Changelog
|
|
| 날짜 | 변경 |
|
|
|---|---|
|
|
| 2026-05-08 | Phase 1 |
|
|
| 2026-05-10 | Manual cleanup — depth cues + stereo rendering + monocular depth estimation |
|