Opencv intro

35
Intro to 影影影影 / 影影 & OpenCV kazami 影影影影影 / 影影影影影

Transcript of Opencv intro

Page 1: Opencv intro

Intro to 影像處理 / 辨識 &

OpenCV kazami前影像處理 / 辨識工程師

Page 2: Opencv intro

我是誰?影像處理 / 辨識工程師大學:化工碩士:理論物理(廣義相對論 \ 重力波 \ 光應用物理 )

Page 3: Opencv intro

所以 ...

影像處理 / 辨識工作到底在幹嘛?

Page 4: Opencv intro

xxx Bank

Page 5: Opencv intro

一開始 , 大家認為影像處理是這樣

https://www.youtube.com/watch?v=csvt6JBAwBk

Page 6: Opencv intro

我以為影像處理是這樣

Page 7: Opencv intro

結果我在做這樣 ... Orz…

Page 8: Opencv intro

影像處理是什麼? 能賺錢嗎?Library工廠品保( PCB, ITO 玻璃 , 面板 , 造紙廠 , 花生 )特殊影像識別系統 ( 保全系統 , 指紋辨識系統 , 自動駕駛 )UI 介面 ( 辨識手勢 ...)

Page 9: Opencv intro

Computer Vision

CameraLightImage ProcessingRecognition

Page 10: Opencv intro

Camera- 劇情總是從出乎意料的地方開始Exposure Time: cmos sensor 接受光的時間ISO : 感光度Aperture : 光圈大小 (影響進光量)Frame Rate : 每秒拍攝的頻率

Page 11: Opencv intro

Camera

Page 12: Opencv intro

Light

Intensity :光的強度 (若光不夠可能會曝光不足)Frequency :光的頻段 (紫外光 可見光 紅外線 X光 微波)Color temperature :色溫 ( RGB 比例的偏重)

Page 13: Opencv intro

CCD 感光

Page 14: Opencv intro

Gray = 0.299 * Red + 0.587 * Green + 0.114 * Blue

Page 15: Opencv intro

以上絕對不是要拖台錢為什麼我用相同的演算法總是辨識錯誤?

為什麼昨天辨識成功今天就辨識不出來?

為什麼找到一堆不是我要的資訊?

Page 16: Opencv intro

Digital Image Processing

Pixel: the minimal element for an image

How to describe the information on a pixel?Color space. RGB, HIS, YCbCr, etc.

Page 17: Opencv intro

Digital Image Processing

Processing for what?EnhancementFeature extractionRecoveryRetouchingCompression

Page 18: Opencv intro

Related Libraries

HalconOpenCVSimpleCVEuresys

Page 19: Opencv intro

OpenCVOpenCV (Open Source Computer Vision Library) is an open source library.

computer visionmachine learning

The library has more than 2500 optimized algorithms.

Page 20: Opencv intro

OpenCVInterfaces:

C++CPythonJavaMATLAB

OS:WindowsLinuxAndroidMac OS

Page 21: Opencv intro

OpenCVModules:

corebasic data structuresfile I/O

imgprocimage filteringgeometrical image transformationscolor space conversion

Videomotion estimationobject tracking

Page 22: Opencv intro

OpenCVModules:

calib3dmultiple-view geometry algorithmsstereo correspondence algorithms

features2dfeature detectorsfeature descriptorsfeature matchers

highguivideo capturingimage and video codecssimple UI

GPU (Cuda)

Page 23: Opencv intro

Image Processing

Blurring convolutionkernel(Average)

Page 24: Opencv intro

FaceDetecttrained with particular object (positive example)arbitrary images of the same size (negative example)boosted: create Strong classifier which is a weighted sum of these weak classifierscascade: resultant classifier consists of several simpler classifiers (stages)

Page 25: Opencv intro

Features20*20 demo picturesFor each feature, it finds the best threshold which will classify the faces to positive and negative

Page 27: Opencv intro

C

CvHaarClassifierCascadeCvHaarStageClassifier

CvHaarClassifierCvHaarFeatur

e

Page 29: Opencv intro

CascadeCascade:

Stage1:Classifier11

Feature11Classifier12

Feature12…..

Stage2:Classifier21

Feature21Classifier22

Feature22…..

Page 31: Opencv intro

Machine Learning

Page 32: Opencv intro
Page 33: Opencv intro

Why miscalculations ?training stagehatglassessmileposture

Page 34: Opencv intro

檢測情境的差異Facebook

品質高自動對焦後製

工業手動對焦手動調光速度快精度高

Page 35: Opencv intro

Q&A