Sysmalloc error with opencv

I'm facing a sysmalloc error using opencv. When I debug I find that the error occurs here:

sm = cv::Mat::zeros(h,w,img.type());

Where h and w are respectively img rows and w cols. I displayed them and they were ok. Here is the whole function:

cv::Mat gsmooth(cv::Mat img,int sigma,int radius,std::string methode)
{
    cv::Mat sm;
    std::vector<double> hcol;
    std::vector<double> hrow;
    if(sigma == NULL)
    {
        sigma =1;
    }
    if(radius == NULL)
    {
        radius = ceil(2.5*sigma);
    }
    if(methode.c_str()==NULL)
    {
        methode ="none";
    }

    if(sigma == 0)
    {
        sm = img;
    }
    else
    {

        hcol= gKernel(2*radius+1,sigma);
        hrow= gKernel(2*radius+1,sigma);
        int h=img.rows;
        int w=img.cols;
        int c=img.channels();
        switch (c)
        {
                case 1:
                     sm= cv::Mat::zeros(h,w,img.type());
                    break;
                case 2:
                     sm= cv::Mat::zeros(h,w,CV_32SC2);
                    break;
                default:
                     sm= cv::Mat::zeros(h,w,CV_32SC3);
        }
        if(!methode.compare(std::string("mirror")))
        {
            cv::Mat mattmp=mirror(img,radius,radius);
            sm=conv2(mattmp,hcol,hrow);

            int ma=img.rows;
            int na=img.cols;
            int nb=hcol.size();
            int mb=hrow.size();
            sm=sm.rowRange((mb-1)/2,(mb-1)/2+ma).colRange((nb-1)/2,(nb-1)/2+na);
        }
        else
        {
            cv::Mat sm_;
            sm_=conv2(img,hcol,hrow);
            int H=sm.rows;
            int W=sm.cols;
            int h=img.rows;
            int w=img.cols;
            int y=ceil(H/h);
            int x=ceil(W/w);
            sm=sm.rowRange(y,y+h).colRange(x,x+w);
        }

    }
    hcol.clear();
    hrow.clear();
    return(sm);

}

The function is call by another .cpp with a cv::Mat img, two int and "mirror" or "none". The arguments seem correct: the img matrix is correctly loaded and the other are only int.

The full error is: *malloc.c:2372: sysmalloc: Assertion '(old_top== (((mbinptr) (((char *) &((av)->bins[((1) - 1)*2)) -__builtin_offsetof (struct malloc_chunk, fd)))) && old_size == 0) || ((unsigned long) (old_size) >= (unsigned long)((((__builtin_offsetof (struct malloc_chunk,fd_nextsize))+((2*(sizeof(size_t))) -1)) & ~((2*sizeof(size_t))) -1)))&& ((old_top)-> size & 0x1) && ((unsigned long) old_end & pagemask) ==0)' failed.*

I'm on ubuntu 14.04.5 using Qt 5.2, gcc 4.6 and g++ 4.4 and i've to use opencv 2.4.13 due to compatibility issues with braid lib.

I'm sure that my error is something stupid, maybe I can't declare a matrix of zero like this but I don't know how to do otherwise and I didn't find any topic relatives with this issue.

In general I'm quite stuck with malloc issue using opencv. Most of the times, i'm using function getting a cv::Mat in argument and using it to return a new matrix declared inside the function, like this:

cv::Mat function somefunction(cv::Mat img, some arguments)
{
   cv::Mat res;
   //operations 
   return(res);
}

and I call the function like this: new_img=some_function(img, some_arguments);

I wonder if it's the right way to do?

I hope I've been specific enough, don't hesitate if you need more information.

EDIT: Using Valgrind I found that the issue occurs when I try to use gKernel. In fact gsmooth is call two times, the first one everything is ok and the crash occurs during the second call. Valgrind pointed a SIGSEGV error at the first line of the function: std::vector<double> kernel(size);

Here is the entire function:

std::vector<double> gKernel(int size, float sigma)
{
   std::vector<double>kernel(size);
   double r,s =2.0* sigma * sigma;
   double sum=0.0;
   int radius=(size -1)/2;
   int index=0;
   for(int y=-radius; y<=radius;y++)
   {
       r=sqrt(y*y);
       kernel[index]=(exp(-(r*r)/s))/(sqrt(2*M_PI*s));
       sum+=kernel[index];
       index++;
   }
   for(int i=0;i<size;i++)
   {
       kernel[i]/=sum;
   }
   return(kernel);
}

I don't understand why declaring a vector may cause a Segfault? As you can see in my gsmooth function, gKernel is call with size = 2*radius +1 and sigma is equal to 60 in the first call and 50 to the second one.

At least I learned that I shouldn't copy cv::Mat with '=' but rather with clone or copyTo to avoid jumbling up pixels'pointers.