安装

#3.4.2

sudo apt-get install opencv

sudo apt-get install golang-1.9

vi ~/.profile

export GOPATH=$HOME/Desktop/GoPath

export PATH=$PATH:$GOPATH/bin

:wq!

#https://github.com/hybridgroup/gocv

#version 0.14.0

go get gocv.io/x/gocv

gocv

//灰度图

gocv.CvtColor(img,&grayMat, gocv.ColorRGBAToGray)

//均值滤波

gocv.Blur(grayMat, &canny, image.Point{3,3})

//边缘检测

gocv.Canny(grayMat, &canny, 125, 225)

//二值化处理，非黑即白图片

gocv.Threshold(canny, &binaryMat, 165, 255, gocv.ThresholdBinary)

//中值滤波

gocv.MedianBlur(binaryMat,&binaryMat,3)

//创建一个40x40的Mat

element_9 := gocv.NewMatWithSizeFromScalar(gocv.NewScalar(1,1,1,255), 40, 40, gocv.MatTypeCV8U)

defer element_9.Close()

//将图片binaryMat 中 + - 40 距离的连通矩形 连在一起

gocv.MorphologyEx(binaryMat,&element_9,gocv.MorphCross, element_9)

//寻找连通矩形

contoursList := gocv.FindContours(element_9, gocv.RetrievalCComp,gocv.ChainApproxNone)

for _,contours := range(contoursList){

rect := gocv.BoundingRect(contours)

gocv.Rectangle(&img,rect,blue,1)

}

准备数据

mkdir opencv_data

cd opencv_data

#放置训练数据图片(大小一致)

mkdir img

#放置纯背景图片

mkdir neg_img

#根据提示操作，选择需要识别的区域

opencv_annotation --annotations=positive.txt --images=img

#将positive.txt中的图片对应的矩阵放置到 positive_sample.vec中等待训练使用

opencv_createsamples -info positive.txt -vec positive_sample.vec -num 66 -w 100 -h 63

训练数据

mkdir data

opencv_traincascade -data data -vec positive_sample.vec -bg negative.txt -numPos 66 -numNeg 7 -numStage 15 -w 100 -h 63 -minHitRate 0.99 -maxFalseAlarmRate 0.5 -mode ALL -featureType HAAR -precalcValBufSize 4048 -precalcIdxBufSize 4048 -numThreads 24

//Required leaf false alarm rate achieved. Branch training terminated

//样本太少

有疑问加站长微信联系(非本文作者)