estimateProjectionMatrix: consistently use double

This leads to a more precise projection matrix, and, in some cases, avoids high residuals although the point cloud is projectable.
Fixes #6181

common/include/pcl/common/impl/projection_matrix.hpp

@@ -103,18 +103,18 @@ estimateProjectionMatrix (
 
   while (pointIt)
   {
-    unsigned yIdx = pointIt.getCurrentPointIndex () / cloud->width;
-    unsigned xIdx = pointIt.getCurrentPointIndex () % cloud->width;
+    Scalar yIdx = pointIt.getCurrentPointIndex () / cloud->width;
+    Scalar xIdx = pointIt.getCurrentPointIndex () % cloud->width;
 
     const PointT& point = *pointIt;
     if (std::isfinite (point.x))
     {
-      Scalar xx = point.x * point.x;
-      Scalar xy = point.x * point.y;
-      Scalar xz = point.x * point.z;
-      Scalar yy = point.y * point.y;
-      Scalar yz = point.y * point.z;
-      Scalar zz = point.z * point.z;
+      Scalar xx = static_cast<Scalar>(point.x) * static_cast<Scalar>(point.x);
+      Scalar xy = static_cast<Scalar>(point.x) * static_cast<Scalar>(point.y);
+      Scalar xz = static_cast<Scalar>(point.x) * static_cast<Scalar>(point.z);
+      Scalar yy = static_cast<Scalar>(point.y) * static_cast<Scalar>(point.y);
+      Scalar yz = static_cast<Scalar>(point.y) * static_cast<Scalar>(point.z);
+      Scalar zz = static_cast<Scalar>(point.z) * static_cast<Scalar>(point.z);
       Scalar xx_yy = xIdx * xIdx + yIdx * yIdx;
 
       A.coeffRef (0) += xx;