symao / libicp Goto Github PK

Hello, everyone
I compile libicp by MinGW and MATLAB2017a . There is an error:
fatal error: boost/multi_array.hpp: No such file or directory
compilation terminated.
I am looking forward to your help.
Thanks.

Hi, I'm attempting to use this library to align some point clouds. On some clouds, for reasons I've yet to identify, I'm getting crashes in the build_tree_for_Range function, particularly in

` int c = -1;
float maxspread = 0.0;
int m;

  for (int i=0;i<dim;i++) {
    if ((parent == NULL) || (parent->cut_dim == i)) {
      spread_in_coordinate(i, l, u, node->box[i]);
    } else {
      node->box[i] = parent->box[i];
    }
    float spread = node->box[i].upper - node->box[i].lower;
    if (spread>maxspread) {
      maxspread = spread;
      c=i;
    }
  }`

where spread>maxspread never hits, and the array tries to index c=-1, which crashes.

i have a model with 9 points :

[ first model ]

point [1] = -0.7274105,0.4433649,0.3854795

point [2] = -0.6887414,0.468136,0.3351357

point [3] = -0.6507958,0.4968288,0.3064253

point [4] = -0.608408,0.5224559,0.2845465

point [5] = -0.5705151,0.5428149,0.2762859

point [6] = -0.5200281,0.530516,0.290995

point [7] = -0.4738899,0.517576,0.3070973

point [8] = -0.4077983,0.4964231,0.3430573

point [9] = -0.374032,0.4680868,0.3830469

[ Second model ]

point [1] = -0.2999626,-0.1806934,-0.1651699

point [2] = -0.2616181,-0.169532,-0.2044329

point [3] = -0.2160359,-0.1454861,-0.2266136

point [4] = -0.1741856,-0.1275599,-0.2524842

point [5] = -0.1191077,-0.105789,-0.251252

point [6] = -0.05154949,-0.12353,-0.2471861

point [7] = -0.006585903,-0.14095,-0.2418299

point [8] = 0.03595421,-0.1584896,-0.2201264

point [9] = 0.08099623,-0.1683025,-0.189768

we got this result :
Transformation results:
R:
-0.0280522 -0.6967893 0.7167271
0.6285203 0.5452465 0.5546787
-0.7772871 0.4660376 0.4226508

t:
16.1951564
3.4604233
-1.0535334

Why do we got this result, because after we apply it to the models, we still not have accurate matching

#pragma omp parallel for private(i) default(none) shared(T,active,nact,p_m,p_t,r00,r01,r10,r11,t0,t1) reduction(+:mum0,mum1, mut0,mut1) // schedule (dynamic,2)

how to get correspondences p_m and p_t?
if M and T do not have the same point numbers,is it still work?

I tried with small arrays, it worked but it raised error with the larger ones.

The error was as following:

Running ICP (point-to-plane, no outliers)
openGLScratch: /mnt/linuxdisk/localdev/boost_1_76_0/build/include/boost/multi_array/base.hpp:177: Reference boost::detail::multi_array::value_accessor_one<T>::access(boost::type<Reference>, boost::detail::multi_array::value_accessor_one<T>::index, TPtr, const size_type*, const index*, const index*) const [with Reference = const float&; TPtr = const float*; T = float; boost::detail::multi_array::value_accessor_one<T>::index = long int; boost::detail::multi_array::multi_array_base::size_type = long unsigned int]: Assertion `idx - index_bases[0] >= 0' failed.
Aborted (core dumped)

I traced down to the code in kdtree.cpp:185

            sum += the_data[ind[k]][c];

Here are the point cloud files

https://www.dropbox.com/sh/7ov8hf5dp45d42r/AACDvZlPWiziFxkkSjmEZdpJa?dl=0

abs() method should be used here:

Hi,

I was using the point to plane ICP on 2D data. While the results are correct when I use Eigen::umeyama(), the results I get when I run the IcpPointToPlane gives
Transformation results:
R:
1.0000000 0.0000000 0.0000000
0.0000000 1.0000000 0.0000000
0.0000000 0.0000000 1.0000000

t:
0.0000000
0.0000000
0.0000000

I am testing it with a simple function y = x² + c on 50 points and 1000 points.
I just changed the demo.cpp as follows

int32_t dim = 2;
int32_t num = 1000;

// allocate model and template memory
double* M = (double*)calloc(dimnum,sizeof(double));
double T = (double*)calloc(dim*num,sizeof(double));

// set model and template points
cout << endl << "Creating model with 10000 points ..." << endl;
cout << "Creating template by shifting model by (1,0.5,-1) ..." << endl;
int32_t k=0;

for(int i=0; i<num; i++){
M[k * dim+0] = i;
M[k * dim+1] = std::pow(i,2);
T[k * dim+0] = i;
T[k * dim+1] = std::pow(i,2)+10;
k++;
}

// start with identity as initial transformation
// in practice you might want to use some kind of prediction here
Matrix R = Matrix::eye(dim);
Matrix t(dim,1);

// run point-to-plane ICP (-1 = no outlier threshold)
cout << endl << "Running ICP (point-to-plane, no outliers)" << endl;
IcpPointToPlane icp(M,num,dim);
double residual = icp.fit(T,num,R,t,-1);

I am not sure what I am missing. Eigen::umeyama() gives the correct translation of (0,10) for the 2 curves.

Thanks!