Overview of Ball Calibration

The ball calibration process was developed using a ball, achieving fast and accurate multi-camera calibration. For details, you can also refer to multiview_ball_calib. Here, we only present the general process.

Why Use a Ball for Calibration

Calculating the extrinsic parameters (position and rotation angle) of multiple cameras is essential for establishing an accurate 3D coordinate model. Traditional checkerboard extrinsic calibration methods require multiple cameras to simultaneously capture the front pattern of the checkerboard, but in a surround multi-camera configuration, there are viewing angle limitations. The application efficiency of checkerboards is low, while using a ball is more efficient.

Experimental Procedure

Create a metal target ball with a diameter of 1.5 cm, suspended by a short rod. Before the experiment, place a checkerboard on the ground at the center of the recording environment to align the coordinate system. Initially, place the target ball away from the checkerboard to avoid occlusion. Then, start recording and traverse the ball around the rat's activity space for 2 minutes. Use OBS Studio software to capture multi-view videos, ensuring the ball's motion trajectory covers the spatial positions that need calibration.

Video Analysis with Code

#!/bin/bash
conda activate mmdet
vfile='/PATH_to_ball/BALL____2024-01-22_14-34-01.mp4'
LILAB_DIR=/home/liying_lab/chenxf/ml-project/LILAB-py/lilab

bash $LILAB_DIR/multiview_scripts_dev/p_calibration.sh $vfile carl

Where vfile is the video file path recorded by OBS Studio software, and carl is the camera calibration parameter. The script will automatically analyze the video and generate calibration results. Common camera calibration parameters include: - ana: Rat Behavior Room 1 - bob: Rat Behavior Room 2 / Small Bucket - carl: Rat Behavior Room 3 / Medium & Large Bucket.

Calibration Result Display

Check the green prediction points' fit with the ball in the generated *ball_keypoint.mp4. The generated *.calibpkl is the multi-camera model file used for subsequent 3D reconstruction.

Calibration Accuracy Evaluation

Check the code execution output. The Reprojection errors represent the reprojection error, which is the distance between the predicted point and the actual point. The smaller the error, the higher the calibration accuracy. Typically, it is allowed within 4.0+-2.0 pixels.