Implement ball diameter-based calibration for distance and 3D position

[atharva-matale]

Closes #5

This PR implements the auto-calibration features for distance and 3D position estimation using the known golf ball diameter. The estimate_distance and estimate_3d_position functions in calibration.rs now utilize the detected ball radius and intrinsic camera parameters to calculate these values, removing the previous TODO stubs and improving accuracy. Updates were also made to trajectory.rs to utilize the new calibration outputs.

[r87-e]

Thanks for tackling this! The approach is right but there are a few issues to fix:

pixel_to_world() — sensor pixel size bug

let distance_mm = (fx * real_diameter_mm) / (radius_px * 0.001);

The hardcoded 0.001 (1μm pixel size) is wrong — the IMX296 has 3.45μm pixels. But more importantly, this formula mixes focal length in mm with pixel coordinates. The standard pinhole model is:

distance = (focal_length_px * real_diameter_mm) / (2 * radius_px)

Where focal_length_px is already in self.intrinsic_matrix[0] (fx). No sensor pixel size needed.

auto_calibrate() — circular logic

You read self.intrinsic_matrix[0] to compute focal length, then write it back to self.intrinsic_matrix[0]. On the first call with the default value this might produce a reasonable number, but subsequent calls would drift. The formula should use ball_radius_px and known_distance_mm directly:

focal_length_px = (2 * ball_radius_px * known_distance_mm) / real_diameter_mm

Missing

• No tests — please add at least one test for pixel_to_world with known values
• Missing newline at end of file

The math is close, just needs these corrections. Looking forward to the update!