diff --git a/modules/core/src/parallel_pthreads.cpp b/modules/core/src/parallel_pthreads.cpp
index f46515f37a..8c34959783 100644
--- a/modules/core/src/parallel_pthreads.cpp
+++ b/modules/core/src/parallel_pthreads.cpp
@@ -252,7 +252,14 @@ pthread_mutex_t ThreadManager::m_manager_access_mutex = PTHREAD_RECURSIVE_MUTEX_
 
 ThreadManager::ptr_holder ThreadManager::m_instance;
 const char ThreadManager::m_env_name[] = "OPENCV_FOR_THREADS_NUM";
+
+#ifdef ANDROID
+// many modern phones/tables have 4-core CPUs. Let's use no more
+// than 2 threads by default not to overheat the devices
+const unsigned int ThreadManager::m_default_number_of_threads = 2;
+#else
 const unsigned int ThreadManager::m_default_number_of_threads = 8;
+#endif
 
 ForThread::~ForThread()
 {
@@ -399,11 +406,10 @@ ThreadManager::~ThreadManager()
 
 void ThreadManager::run(const cv::Range& range, const cv::ParallelLoopBody& body, double nstripes)
 {
-    bool is_work_thread;
-
-    is_work_thread = m_is_work_thread.get()->value;
+    bool is_work_thread = m_is_work_thread.get()->value;
 
-    if( (getNumOfThreads() > 1) && !is_work_thread && (range.end - range.start > 1) )
+    if( (getNumOfThreads() > 1) && !is_work_thread &&
+        (range.end - range.start > 1) && (nstripes <= 0 || nstripes >= 1.5) )
     {
         int res = pthread_mutex_trylock(&m_manager_access_mutex);
 
diff --git a/modules/video/src/ecc.cpp b/modules/video/src/ecc.cpp
index d527f8f890..cc08ac005c 100644
--- a/modules/video/src/ecc.cpp
+++ b/modules/video/src/ecc.cpp
@@ -402,16 +402,19 @@ double cv::findTransformECC(InputArray templateImage,
 
     Mat inputMaskMat = inputMask.getMat();
     //to use it for mask warping
-    Mat preMask =
-      inputMaskMat.empty() ? Mat::ones(hd, wd, CV_8U) : inputMaskMat;
+    Mat preMask;
+    if(inputMask.empty())
+        preMask = Mat::ones(hd, wd, CV_8U);
+    else
+        threshold(inputMask, preMask, 0, 1, THRESH_BINARY);
 
     //gaussian filtering is optional
     src.convertTo(templateFloat, templateFloat.type());
-    GaussianBlur(templateFloat, templateFloat, Size(5, 5), 0, 0);//is in-place filtering slower?
+    GaussianBlur(templateFloat, templateFloat, Size(5, 5), 0, 0);
 
-    Mat preMaskFloat = Mat(hd, wd, CV_32F);
-    preMask.convertTo(preMaskFloat, preMaskFloat.type());
-    GaussianBlur(preMaskFloat, preMaskFloat, Size(5, 5), 0, 0);//is in-place filtering slower?
+    Mat preMaskFloat;
+    preMask.convertTo(preMaskFloat, CV_32F);
+    GaussianBlur(preMaskFloat, preMaskFloat, Size(5, 5), 0, 0);
     // Change threshold.
     preMaskFloat *= (0.5/0.95);
     // Rounding conversion.
@@ -512,7 +515,7 @@ double cv::findTransformECC(InputArray templateImage,
         // calculate enhanced correlation coefficiont (ECC)->rho
         last_rho = rho;
         rho = correlation/(imgNorm*tmpNorm);
-        if (isnan(rho)) {
+        if (cvIsNaN(rho)) {
           CV_Error(Error::StsNoConv, "NaN encountered.");
         }