diff --git a/cmake/OpenCVDetectCUDA.cmake b/cmake/OpenCVDetectCUDA.cmake
index 58e204094f..27bf79bd8e 100644
--- a/cmake/OpenCVDetectCUDA.cmake
+++ b/cmake/OpenCVDetectCUDA.cmake
@@ -70,6 +70,12 @@ if(CUDA_FOUND)
     unset(CUDA_ARCH_PTX CACHE)
   endif()
 
+  SET(DETECT_ARCHS_COMMAND "${CUDA_NVCC_EXECUTABLE}" ${CUDA_NVCC_FLAGS} "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run")
+  if(WIN32 AND CMAKE_LINKER) #Workaround for VS cl.exe not being in the env. path
+    get_filename_component(host_compiler_bindir ${CMAKE_LINKER} DIRECTORY)
+    SET(DETECT_ARCHS_COMMAND ${DETECT_ARCHS_COMMAND} "-ccbin" "${host_compiler_bindir}")
+  endif()
+
   set(__cuda_arch_ptx "")
   if(CUDA_GENERATION STREQUAL "Fermi")
     set(__cuda_arch_bin "2.0")
@@ -82,10 +88,11 @@ if(CUDA_FOUND)
   elseif(CUDA_GENERATION STREQUAL "Volta")
     set(__cuda_arch_bin "7.0")
   elseif(CUDA_GENERATION STREQUAL "Auto")
-    execute_process( COMMAND "${CUDA_NVCC_EXECUTABLE}" ${CUDA_NVCC_FLAGS} "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run"
+    execute_process( COMMAND ${DETECT_ARCHS_COMMAND}
                      WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
                      RESULT_VARIABLE _nvcc_res OUTPUT_VARIABLE _nvcc_out
                      ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
+    string(REGEX REPLACE ".*\n" "" _nvcc_out "${_nvcc_out}") #Strip leading warning messages, if any
     if(NOT _nvcc_res EQUAL 0)
       message(STATUS "Automatic detection of CUDA generation failed. Going to build for all known architectures.")
     else()
@@ -99,10 +106,11 @@ if(CUDA_FOUND)
       set(__cuda_arch_bin "3.2")
       set(__cuda_arch_ptx "")
     elseif(AARCH64)
-      execute_process( COMMAND "${CUDA_NVCC_EXECUTABLE}" ${CUDA_NVCC_FLAGS} "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run"
+      execute_process( COMMAND ${DETECT_ARCHS_COMMAND}
                        WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
                        RESULT_VARIABLE _nvcc_res OUTPUT_VARIABLE _nvcc_out
                        ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
+      string(REGEX REPLACE ".*\n" "" _nvcc_out "${_nvcc_out}") #Strip leading warning messages, if any
       if(NOT _nvcc_res EQUAL 0)
         message(STATUS "Automatic detection of CUDA generation failed. Going to build for all known architectures.")
         set(__cuda_arch_bin "5.3 6.2 7.0")
diff --git a/cmake/checks/OpenCVDetectCudaArch.cu b/cmake/checks/OpenCVDetectCudaArch.cu
index 9d7086cf24..70ca975530 100644
--- a/cmake/checks/OpenCVDetectCudaArch.cu
+++ b/cmake/checks/OpenCVDetectCudaArch.cu
@@ -1,14 +1,25 @@
-#include <stdio.h>
+#include <iostream>
+#include <sstream>
+#include <list>
+
 int main()
 {
+    std::ostringstream arch;
+    std::list<std::string> archs;
+
     int count = 0;
-    if (cudaSuccess != cudaGetDeviceCount(&count)){return -1;}
-    if (count == 0) {return -1;}
+    if (cudaSuccess != cudaGetDeviceCount(&count)){ return -1; }
+    if (count == 0) { return -1; }
     for (int device = 0; device < count; ++device)
     {
         cudaDeviceProp prop;
-        if (cudaSuccess != cudaGetDeviceProperties(&prop, device)){ continue;}
-        printf("%d.%d ", prop.major, prop.minor);
+        if (cudaSuccess != cudaGetDeviceProperties(&prop, device)){ continue; }
+        arch << prop.major << "." << prop.minor;
+        archs.push_back(arch.str());
+        arch.str("");
     }
+    archs.unique(); #Some devices might have the same arch
+    for (std::list<std::string>::iterator it=archs.begin(); it!=archs.end(); ++it)
+        std::cout << *it << " ";
     return 0;
 }