Stereo & Depth Estimation¶

stereo_3d_reconstruction: Depth-based point cloud analysis: colorize by depth, filter by depth range. Foundation for stereo vision and 3D reconstruction from camera images.


Methods	`depth_colorization`, `depth_filter`
Input files	`.ply`, `.pcd`, `.txt`, `.xyz`
Bundled datasets to try	`open3d_bunny`, `open3d_icp_demo`, `kitti_raw_2011_09_26_drive_0001`, `medshapenet`, `c3vd_endoscopy_splats`
CLI	`pardesline run stereo_3d_reconstruction.<method> input.ext [--param value]`

How it works¶

Stereo vision reconstructs 3D structure from camera views. The fundamental relationship between depth $Z$, disparity $d$, focal length $f$, and baseline $B$ is:

\[ Z = \frac{f \cdot B}{d} \]

A 3D point $\mathbf{X} = (X, Y, Z)^T$ in world coordinates projects to pixel $(u, v)$ via the camera intrinsic matrix $K$:

\[ \begin{pmatrix} u \\ v \\ 1 \end{pmatrix} = \frac{1}{Z} K \mathbf{X}, \quad K = \begin{pmatrix} f_x & 0 & c_x \\ 0 & f_y & c_y \\ 0 & 0 & 1 \end{pmatrix} \]

For point cloud analysis, depth-based operations include colorization by Euclidean distance and depth-range filtering for region-of-interest extraction.

Methods¶

depth_colorization¶

1. Distance Computation For each point $p_i$ in the cloud, the Euclidean distance from the origin (camera center) is computed: $$d_i = \|p_i - \mathbf{o}\|_2 = \sqrt{x_i^2 + y_i^2 + z_i^2}$$

2. Normalization and Colormap Distances are normalized to $[0, 1]$ and mapped through a perceptually uniform colormap (jet, viridis, plasma, or turbo): $$c_i = \text{colormap}\left(\frac{d_i - d_{\min}}{d_{\max} - d_{\min}}\right)$$

import pardesline as pl

obj = pl.load("input.ply")
result = obj.run("stereo_3d_reconstruction.depth_colorization")  # catalog defaults
result.provenance[-1]  # the step, every parameter used, duration, version

Output: .ply. Script: python -m pardesline.ops.stereo_3d_reconstruction.depth_colorization --input ... --output ....

Defaults validated on: generated depth clouds in tests/test_ops_smoke.py

Parameter	Type	Default	Range	What it does	Why this default, when to change it
`colormap`	int	`0`	0 to 3	Colormap: 0=jet, 1=viridis, 2=plasma, 3=turbo	0 (jet) is what stereo and depth tooling has used for decades, so the colours are readable at a glance; the others are perceptually uniform alternatives.

depth_filter¶

1. Percentile-Based Filtering Instead of absolute depth thresholds, the filter uses percentiles of the distance distribution for robustness across different scales: $$P_{\text{keep}} = \{p_i : q_{\min} \leq d_i \leq q_{\max}\}$$ where $q_{\min}, q_{\max}$ are the min_depth and max_depth percentiles of $\{d_i\}$.

2. Application Useful for removing background points (far range) or ground plane (near range) in LiDAR and depth camera data.

import pardesline as pl

obj = pl.load("input.ply")
result = obj.run("stereo_3d_reconstruction.depth_filter")  # catalog defaults
result.provenance[-1]  # the step, every parameter used, duration, version

Output: .ply. Script: python -m pardesline.ops.stereo_3d_reconstruction.depth_filter --input ... --output ....