Emgu(opencv)利用Tensorflow识别结合MatchTemplate匹配图像准确判别船舶运动方向
项目知识点记录
1、利用tensorflow训练模型让其识别船舶
2、得到识别的物体矩形坐标后,通过下面方法提取感兴趣区域图像
/// <summary>
/// 安装矩形裁剪指定图片
/// </summary>
/// <param name="est"></param>
/// <param name="rectangle"></param>
/// <returns></returns>
Image<Bgr, byte> CutRectangularPictures( Image<Bgr, byte> est, Rectangle rectangle)
{
Image<Bgr, byte> Sub = est;
//Rectangle rectangle = new Rectangle(240, 340, 300, 100);
CvInvoke.cvGetSubRect(est, Sub,rectangle);
Image<Bgr, byte> CropImage = new Image<Bgr, byte>(Sub.Size);
CvInvoke.cvCopy(Sub, CropImage, IntPtr.Zero);
return CropImage;
}
3、利用2得到我们感兴趣的区域后,再执行下面函数可以匹配到感兴趣区域在原图的位置
/// <summary>
/// 找到匹配的图像
/// </summary>
/// <param name="a"></param>
/// <param name="b"></param>
/// <returns></returns>
Image<Bgr, byte> FindingMatchedImageTracking(Image<Bgr, byte> a, Image<Bgr, byte> b)
{
//= new Image<Bgr, byte>(@"D:\1.png");
// = new Image<Bgr, byte>(@"D:\2.png");
Image<Gray, float> c = new Image<Gray, float>(a.Width, a.Height);
c = a.MatchTemplate(b, TemplateMatchingType.Ccoeff);
forgroundImageBox.Image = c;
double min = 0, max = 0;
Point maxp = new Point(0, 0);
Point minp = new Point(0, 0);
CvInvoke.MinMaxLoc(c, ref min, ref max, ref minp, ref maxp);
Console.WriteLine(min + ' ' + max);
//CvInvoke.Normalize(c, c);
CvInvoke.Rectangle(a, new Rectangle(maxp, new Size(b.Width, b.Height)), new MCvScalar(0, 0, 255), 3);
return a;
}
4、运行效果GIF图
可以看到感兴趣部分很稳的跟踪着匹配图像的上