点云处理

记录一些点云处理相关的东西。

• 注意检查函数的输入是弧度还是角度

• 不同软件使用的坐标系

  

  调用不同的库之前要记得进行坐标系的变换。

• 深度图不能resize太小，否则转成点云后，物体边缘的点会不准。

• pytorch实现voxel_downsample，在4090上百万个点降采样只要几毫秒。

    import torch
  
    def unique(x, dim=None):
        """Unique elements of x and indices of those unique elements
        https://github.com/pytorch/pytorch/issues/36748#issuecomment-619514810
  
        e.g.
  
        unique(tensor([
            [1, 2, 3],
            [1, 2, 4],
            [1, 2, 3],
            [1, 2, 5]
        ]), dim=0)
        => (tensor([[1, 2, 3],
                    [1, 2, 4],
                    [1, 2, 5]]),
            tensor([0, 1, 3]))
        """
        unique, inverse = torch.unique(x, sorted=True, return_inverse=True, dim=dim)
        perm = torch.arange(inverse.size(0), dtype=inverse.dtype, device=inverse.device)
        inverse, perm = inverse.flip([0]), perm.flip([0])
        index = inverse.new_empty(unique.size(0)).scatter_(0, inverse, perm)
        return unique, index
  
    def voxel_downsample(points, colors, features=None, cell_size=0.1):
        points_ = torch.round(points / cell_size)
        points_down, index = unique(points_, dim=0)
        points_down = points_down * cell_size
        colors_down = colors[index, :]
        if features is None:
            return points_down, colors_down
        features_down = features[index, :]
        return points_down, colors_down, features_down