Functions
Image< size_t >	downsample_depth (const Image< float > &input_depth_img, Image< float > &output_depth_img)
	Perform median downsampling on `input_depth_img` and save the result in `output_depth_img`.

template<typename SensorT >
void	depth_to_point_cloud (se::Image< Eigen::Vector3f > &point_cloud_C, const se::Image< float > &depth_image, const SensorT &sensor)

void	point_cloud_to_depth (se::Image< float > &depth_image, const se::Image< Eigen::Vector3f > &point_cloud_X, const Eigen::Isometry3f &T_CX)

template<bool NegY>
void	point_cloud_to_normal (se::Image< Eigen::Vector3f > &out, const se::Image< Eigen::Vector3f > &in)
	NegY should only be true when reading an ICL-NUIM dataset which has a left-handed coordinate system (the y focal length will be negative).

void	half_sample_robust_image (se::Image< float > &out, const se::Image< float > &in, const float e_d, const int r)

Function Documentation

◆ downsample_depth()

Image< size_t > se::preprocessor::downsample_depth	(	const Image< float > &	input_depth_img,
		Image< float > &	output_depth_img
	)

Perform median downsampling on input_depth_img and save the result in output_depth_img.

Return an image with the same dimensions as output_depth_img containing linear indices to input_depth_img. It can be used with se::image::remap() to downsample another image (e.g. colour) and select the same pixels as the ones used for depth downsampling.

Median downsampling avoids introducing new, non-measured depth values, which can create artifacts near object edges. The median of an even number of depth values is computed as the smallest of the two middle depth values instead of their mean which would introduce a new, non-measured depth value. Depth values of 0 or NaN are considered invalid and are ignored when computing the median.

Note: The ratio between the input and output image resolutions must be a power of two.

◆ depth_to_point_cloud()

template<typename SensorT >

void se::preprocessor::depth_to_point_cloud	(	se::Image< Eigen::Vector3f > &	point_cloud_C,
		const se::Image< float > &	depth_image,
		const SensorT &	sensor
	)

◆ point_cloud_to_depth()

void se::preprocessor::point_cloud_to_depth	(	se::Image< float > &	depth_image,
		const se::Image< Eigen::Vector3f > &	point_cloud_X,
		const Eigen::Isometry3f &	T_CX
	)

◆ point_cloud_to_normal()

template<bool NegY>

void se::preprocessor::point_cloud_to_normal	(	se::Image< Eigen::Vector3f > &	out,
		const se::Image< Eigen::Vector3f > &	in
	)

NegY should only be true when reading an ICL-NUIM dataset which has a left-handed coordinate system (the y focal length will be negative).

◆ half_sample_robust_image()