学步园

短焦镜头通常会产生桶形失真，以下是校正算法的matlab代码：

%% 镜头桶形失真校正(短焦镜头)</p> <p>img_origin1 = imread('Still001.bmp');<br /> img_origin = rgb2gray(img_origin1);</p> <p>k1 = -0.00000037; % 形变参数，根据实际情况调整<br /> k2 = -0.00000037;</p> <p>img_size = size( img_origin );<br /> img_undist = zeros( img_size );<br /> img_undist = uint8( img_undist );</p> <p>for l1 = 1:img_size(1) % 垂直方向</p> <p> y = l1 - img_size(1)/2;</p> <p> for l2 = 1:img_size(2) % 水平方向</p> <p> x = l2 - img_size(2)/2;</p> <p> x1 = round( x * ( 1 + k1 * x * x + k2 * y * y ) );<br /> y1 = round( y * ( 1 + k1 * x * x + k2 * y * y ) );<br /> y1 = y1 + img_size(1)/2;<br /> x1 = x1 + img_size(2)/2;</p> <p> img_undist(l1,l2) = img_origin(y1, x1);</p> <p> end<br /> end</p> <p>figure(1);<br /> subplot(121); imshow(img_origin);<br /> subplot(122); imshow(img_undist);</p> <p>imwrite(img_origin,'1.bmp');<br /> imwrite(img_undist,'2.bmp');

在实时视频处理中，如果对每帧图像都要重复进行坐标转换运算，就是个很大的浪费，我们可以在初始化过程中先将坐标对应关系建立一个二维查找表，之后每帧处理只需要查表，C++代码如下（用到了OpenCV）：

// 镜头校正查找表<br /> CvPoint CorrInd[480][640];</p> <p>// 镜头径向畸变校正初始化<br /> void LensCorrectInit( double k1 = -0.00000035, double k2 = -0.00000035 );</p> <p>// 镜头径向畸变校正初始化<br /> void LensCorrectInit( double k1, double k2 )<br /> {<br /> int imgWidth = camera.GetWidth(); // 获得摄像头图像尺寸<br /> int imgHeight = camera.GetHeight();</p> <p> int imgWd2 = imgWidth>>1;<br /> int imgHd2 = imgHeight>>1;</p> <p> int it;<br /> int jt;<br /> int it2;<br /> int jt2;</p> <p> for( int i=0; i<imgHeight; i++ )<br /> {<br /> it = ( i + 1 ) - imgHd2; // (i+1) 是其真正的图像坐标<br /> for( int j=0; j<imgWidth; j++ )<br /> {<br /> double temp;</p> <p> jt = ( j + 1 ) - imgWd2; // (j+1) 是其真正的图像坐标<br /> temp = 1 + k1*it*it + k2*jt*jt;<br /> it2 = (int)(temp * it);<br /> jt2 = (int)(temp * jt);</p> <p> it2 = it2 + imgHd2 + 1;<br /> jt2 = jt2 + imgWd2 + 1;</p> <p> CorrInd[i][j].x = jt2;<br /> CorrInd[i][j].y = it2;<br /> }<br /> }<br /> }</p> <p>// 镜头径向畸变校正<br /> void LensCorrect( IplImage* src, IplImage* dst )<br /> {<br /> int imgWidth = dst->width;<br /> int imgHeight = dst->height;</p> <p> int it2;<br /> int jt2;</p> <p> for( int i=0; i<imgHeight; i++ )<br /> {<br /> uchar *pdstdata = (uchar*)(dst->imageData) + i*dst->widthStep;</p> <p> for( int j=0; j<imgWidth; j++ )<br /> {<br /> it2 = CorrInd[i][j].y;<br /> jt2 = CorrInd[i][j].x;</p> <p> uchar * psrcdata = cvPtr2D( src, it2, jt2 );</p> <p> *(pdstdata++) = *(psrcdata++);<br /> *(pdstdata++) = *(psrcdata++);<br /> *(pdstdata++) = *(psrcdata++);<br /> }<br /> }<br /> }