Android使用OpenCV CamShift实现目标追踪

CamShift算法基于色值，适用于追踪颜色和背景差异较大的目标。

效果图

以下调试代码，仅供参考：

源码

package com.kongqw; import android.graphics.Bitmap; import org.opencv.android.Utils; import org.opencv.core.Core; import org.opencv.core.CvType; import org.opencv.core.Mat; import org.opencv.core.MatOfFloat; import org.opencv.core.MatOfInt; import org.opencv.core.Rect; import org.opencv.core.RotatedRect; import org.opencv.core.Scalar; import org.opencv.core.TermCriteria; import org.opencv.imgproc.Imgproc; import org.opencv.video.Video; import java.util.Collections; import java.util.List; import java.util.Vector; /** * Created by kongqingwei on 2017/4/26. * ObjectTracker */ public abstract class ObjectTracker { private Mat hsv, hue, mask, prob; private Rect trackRect; private RotatedRect rotatedRect; private Mat hist; private List<Mat> hsvList, hueList; private Bitmap bitmap; private MatOfFloat ranges; public abstract void onCalcBackProject(Bitmap prob); public ObjectTracker(Mat rgba) { hist = new Mat(); trackRect = new Rect(); rotatedRect = new RotatedRect(); hsvList = new Vector<>(); hueList = new Vector<>(); hsv = new Mat(rgba.size(), CvType.CV_8UC3); mask = new Mat(rgba.size(), CvType.CV_8UC1); hue = new Mat(rgba.size(), CvType.CV_8UC1); prob = new Mat(rgba.size(), CvType.CV_8UC1); bitmap = Bitmap.createBitmap(prob.width(), prob.height(), Bitmap.Config.ARGB_8888); ranges = new MatOfFloat(0f, 256f); } public Bitmap createTrackedObject(Mat mRgba, Rect region) { //将rgb摄像头帧转化成hsv空间的 rgba2Hsv(mRgba); updateHueImage(); Mat tempMask = mask.submat(region); // MatOfFloat ranges = new MatOfFloat(0f, 256f); MatOfInt histSize = new MatOfInt(25); List<Mat> images = Collections.singletonList(hueList.get(0).submat(region)); Imgproc.calcHist(images, new MatOfInt(0), tempMask, hist, histSize, ranges); Bitmap bitmap = Bitmap.createBitmap(hue.width(), hue.height(), Bitmap.Config.ARGB_8888); Utils.matToBitmap(hue, bitmap); // 将hist矩阵进行数组范围归一化，都归一化到0~255 Core.normalize(hist, hist, 0, 255, Core.NORM_MINMAX); trackRect = region; return bitmap; } private void rgba2Hsv(Mat rgba) { Imgproc.cvtColor(rgba, hsv, Imgproc.COLOR_RGB2HSV); //inRange函数的功能是检查输入数组每个元素大小是否在2个给定数值之间，可以有多通道,mask保存0通道的最小值，也就是h分量 //这里利用了hsv的3个通道，比较h,0~180,s,smin~256,v,min(vmin,vmax),max(vmin,vmax)。如果3个通道都在对应的范围内，则 //mask对应的那个点的值全为1(0xff)，否则为0(0x00). int vMin = 65, vMax = 256, sMin = 55; Core.inRange( hsv, new Scalar(0, sMin, Math.min(vMin, vMax)), new Scalar(180, 256, Math.max(vMin, vMax)), mask ); } private void updateHueImage() { hsvList.clear(); hsvList.add(hsv); // hue初始化为与hsv大小深度一样的矩阵，色调的度量是用角度表示的，红绿蓝之间相差120度，反色相差180度 hue.create(hsv.size(), hsv.depth()); hueList.clear(); hueList.add(hue); MatOfInt from_to = new MatOfInt(0, 0); // 将hsv第一个通道(也就是色调)的数复制到hue中，0索引数组 Core.mixChannels(hsvList, hueList, from_to); } public RotatedRect objectTracking(Mat mRgba) { rgba2Hsv(mRgba); updateHueImage(); // 计算直方图的反投影。 // Imgproc.calcBackProject(hueList, new MatOfInt(0), hist, prob, ranges, 255); Imgproc.calcBackProject(hueList, new MatOfInt(0), hist, prob, ranges, 1.0); // 计算两个数组的按位连接（dst = src1 & src2）计算两个数组或数组和标量的每个元素的逐位连接。 Core.bitwise_and(prob, mask, prob, new Mat()); // 追踪目标 rotatedRect = Video.CamShift(prob, trackRect, new TermCriteria(TermCriteria.EPS, 10, 1)); // 将本次最终到的目标作为下次追踪的对象 trackRect = rotatedRect.boundingRect(); rotatedRect.angle = -rotatedRect.angle; Imgproc.rectangle(prob, trackRect.tl(), trackRect.br(), new Scalar(255, 255, 0, 255), 6); Utils.matToBitmap(prob, bitmap); onCalcBackProject(bitmap); return rotatedRect; } }

使用部分

public Mat onCameraFrame(CvCameraViewFrame inputFrame) { mRgba = inputFrame.rgba(); mGray = inputFrame.gray(); if (null == objectTracker) { objectTracker = new ObjectTracker(mRgba) { @Override public void onCalcBackProject(final Bitmap prob) { MainActivity.this.runOnUiThread(new Runnable() { @Override public void run() { imageView.setImageBitmap(prob); } }); } }; } if (null != mTrackWindow) { Log.i(TAG, "onCameraFrame: objectTracker = " + objectTracker + " mTrackWindow = " + mTrackWindow); RotatedRect rotatedRect = objectTracker.objectTracking(mRgba); Imgproc.ellipse(mRgba, rotatedRect, FACE_RECT_COLOR, 6); Rect rect = rotatedRect.boundingRect(); Imgproc.rectangle(mRgba, rect.tl(), rect.br(), FACE_RECT_COLOR, 3); } // System.gc(); return mRgba; }

int xDown; int yDown; @Override public boolean onTouch(View v, MotionEvent event) { int cols = mRgba.cols(); int rows = mRgba.rows(); int xOffset = (mOpenCvCameraView.getWidth() - cols) / 2; int yOffset = (mOpenCvCameraView.getHeight() - rows) / 2; switch (event.getAction()) { case MotionEvent.ACTION_DOWN: xDown = (int) event.getX() - xOffset; yDown = (int) event.getY() - yOffset; break; case MotionEvent.ACTION_UP: int xUp = (int) event.getX() - xOffset; int yUp = (int) event.getY() - yOffset; // 获取跟踪目标 mTrackWindow = new Rect(Math.min(xDown, xUp), Math.min(yDown, yUp), Math.abs(xUp - xDown), Math.abs(yUp - yDown)); // 创建跟踪目标 Bitmap bitmap = objectTracker.createTrackedObject(mRgba, mTrackWindow); imageView.setImageBitmap(bitmap); Toast.makeText(getApplicationContext(), "已经选中跟踪目标！", Toast.LENGTH_SHORT).show(); break; default: break; } return true; }

参考

• 【计算机视觉】人脸实时跟踪1——利用CamShift来跟踪人脸

• 目标跟踪–CamShift