コード例 #1
0
ファイル: Curve.py プロジェクト: YQ161916/pymethods
 def get_normal(self):
     k_test = math.approximate_normal(self-self.centroid)
     i = math.normalize(self[:, 0, None] - self.centroid)
     n = math.normalize(
         self[:, self.shape[-1]//4, None] - self.centroid)
     k_test = math.normalize(
         math.cross(i, n))
     j_test = math.cross(k_test, i)
     alpha1 = math.smallest_angle_between_vectors(j_test, n)
     alpha2 = math.smallest_angle_between_vectors(-j_test, n)
     if alpha1 < alpha2:
         return arrays.Vector(math.normalize(k_test))
     else:
         return arrays.Vector(math.normalize(-k_test))
コード例 #2
0
ファイル: Curve.py プロジェクト: YQ161916/pymethods
    def oriented_transport_frames(self, origin, jVector, return_index=True):

        origin_id = np.argmin(math.l2_norm(
            self - arrays.ColumnVector(origin)
        ))

        transportFrames = self.transport_frames()
        ijk_self = arrays.Basis(transportFrames[origin_id])
        jVector = arrays.ColumnVector(jVector).hat
        projected = jVector.project_to_plane(ijk_self[:, -1])
        j_new = math.normalize(
            projected.change_reference_frame(ijk_self)).squeeze()
        z_new = np.array([0, 0, 1])
        ijk_new = arrays.Basis(
            math.normalize(math.cross(j_new, z_new)), j_new, z_new
        )
        newTransportFrames = np.zeros_like(transportFrames)
        for i, frame in enumerate(transportFrames):
            newTransportFrames[i] = arrays.Basis(frame @ ijk_new)
        if return_index:
            return newTransportFrames, origin_id
        else:
            return newTransportFrames
コード例 #3
0
ファイル: Vector.py プロジェクト: YQ161916/pymethods
 def cross(self, vector: np.ndarray) -> np.ndarray:
     """
     """
     return math.cross(self, vector)
コード例 #4
0
def argSortByBasis(array, ref_basis, n_to_check=20):
    """arg_basis_sort [summary]

    For an ndarray describing the points along a closed contour and a ref_basis
    which describes the plane outpud the ids of the sorted points. The points are sorted such that
    the ndarray[0,:] is the point closest to the i-axis and the orientation of rotation
    is determined by the closest point

    Args:
        ndarray (_np.ndarray)  : N x M matrix where N is the number of points
            and M is the total number of dimensions
        ref_basis (_np.ndarray): 3 x 3 basis matrix discribing
        return_ref (bool, optional): return a copy of ndarray in the
            ref_basis frame. Defaults to True.
    Returns:
        _np.ndarray: list of sorted indices
    """
    meanLocation = np.mean(array, axis=-1, keepdims=True)
    ndarray_centered = array - meanLocation
    n_pts = ndarray_centered.shape[-1]
    # ndarray_centered_2d
    A = ref_basis.T @ ndarray_centered
    ids_vector = np.arange(n_pts)[None, :]
    # concatenate A with id vector
    A_concat = np.concatenate([A, ids_vector], axis=0)
    # initialize a vectore to store the sorted
    sorted_ids = np.zeros(A_concat.shape[-1])
    # reference_basis_2d
    ijk = np.identity(3)
    # project A to the i direction
    project_to_x = math.scalar_project(A, ijk[:, 0, None])
    # ids of points moving in the positive direction
    to_use_ids = np.where(project_to_x > 0)[0]
    # subsample ndarray_centered_2d_concat_with_ids
    subset = A_concat[:, to_use_ids]
    # project the subset values to j
    project_to_y = math.scalar_project(subset[0:-1, :], ijk[1, :])
    # locate the key of the point which is closest to the x axis of the subset
    id_of_subset = np.argmin(np.abs(project_to_y))
    # get the id of the starting point
    start_id = int(subset[-1, id_of_subset])
    # store the starting point in the initialesd vector
    sorted_ids[0] = start_id
    # create a copy of the current point
    current_pt = A_concat[0:-1, start_id, None].copy()
    # delete this point from the ndarray_centered_2d concated with an id vector
    copy_A_cont = np.delete(A_concat, start_id, axis=-1)
    counter = 0
    while copy_A_cont.shape[-1] > 0:
        if copy_A_cont.shape[-1]:
            closest_ind = np.argmin(
                np.linalg.norm(copy_A_cont[0:-1] - current_pt, axis=0))
        counter += 1
        assert counter < n_pts*2, \
            'too many iterations something is wrong with the code'
        current_pt = copy_A_cont[0:-1, closest_ind, None]
        sorted_ids[counter] = int(copy_A_cont[-1, closest_ind])
        copy_A_cont = np.delete(copy_A_cont, closest_ind, axis=-1)

    sorted_ndarray_in_ref = A[:, sorted_ids.astype(int)]
    # check if clockwise
    check = 0
    base_vector = math.normalize(sorted_ndarray_in_ref[:, 0, None])

    for i in list(range(n_to_check)):
        test_vector = math.normalize(sorted_ndarray_in_ref[:, i, None])
        check += math.cross(base_vector, test_vector)[-1]

    # if not rotating clockwise then flip order so that
    # points rotate in a clockwise direction
    if check < 0:
        sorted_ids = np.flipud(sorted_ids)
        sorted_ids = np.roll(sorted_ids, 1)

    return np.array([int(i) for i in sorted_ids])
コード例 #5
0
ファイル: Curve.py プロジェクト: YQ161916/pymethods
 def calc_basis(self):
     i = math.normalize(self[:, 0, None] - self.centroid).squeeze()
     k = self.calc_normal()
     j = math.normalize(math.cross(k, i))
     return arrays.Basis(i.squeeze(), j.squeeze(), k.squeeze())