I'm preparing to enhance my project's capability by measuring not only the speed and angle of a fast-moving object but also its depth and potentially rotation. This next phase involves transitioning from a single-camera setup to utilizing stereo vision with two Global Shutter (GS) cameras on a Raspberry Pi.
Background:
My initial setup involved capturing video indoors with a Raspberry Pi Camera (with the IR filter removed and a 120-degree lens attached) under IR illumination. Video capture was achieved using `libcamera-vid` with high FPS configurations, all processed on a Raspberry Pi 4 with 8GB RAM running Bullseye. I've experimented with various resolutions and FPS rates to optimize capture settings, such as using a 1440x320 resolution at 193 FPS.
Current Objective:
To accurately track the object's 3D trajectory, I'm exploring the use of two GS cameras configured for synchronous capture on a Raspberry Pi. This setup aims to provide a detailed time series of the object's XYZ coordinates and, if possible, its rotation.
Technical Details:
- The cameras are configured using `media-ctl` and `libcamera-vid` for high FPS capture.
- For stereo vision, I plan to use two GS cameras, synchronized frame capture with calibrated cameras, and post processing with opencv
- Post-processing will align the synchronized frames (since the master video may have more frames) using the PTS timecodes; I will ultimately be focusing on a brief but critical window of 5-7 frames for analysis. (so I'm not concerned with the master taking extra frames initially)
![Image]()
Request for Feedback:
I'm currently outlining the hardware and wiring design for synchronous capture with the dual-camera setup. Given my primary experience lies in software and OpenCV analytics, I'm seeking advice, corrections, or insights on the hardware aspect and the overall design for achieving accurate stereo capture.
Thank you for any guidance or suggestions, especially regarding the setup for synchronous capture and any considerations I might have overlooked.
Background:
My initial setup involved capturing video indoors with a Raspberry Pi Camera (with the IR filter removed and a 120-degree lens attached) under IR illumination. Video capture was achieved using `libcamera-vid` with high FPS configurations, all processed on a Raspberry Pi 4 with 8GB RAM running Bullseye. I've experimented with various resolutions and FPS rates to optimize capture settings, such as using a 1440x320 resolution at 193 FPS.
Code:
media-ctl -d /dev/media0 --set-v4l2 "'imx296 10-001a':0 [fmt:SBGGR10_1X10/1440x320 crop:(8,384)/1440x320]"Code:
libcamera-vid --level 4.2 --circular 1 --inline --width 1440 --height 320 --framerate 193 --shutter 100 --gain 16.0 --brightness 0.0 --contrast 1.0 --saturation 0.0 --sharpness 1.0 --denoise cdn_off --save-pts /dev/shm/tst.pts -t 0 -o /dev/shm/output.h264 -nTo accurately track the object's 3D trajectory, I'm exploring the use of two GS cameras configured for synchronous capture on a Raspberry Pi. This setup aims to provide a detailed time series of the object's XYZ coordinates and, if possible, its rotation.
Technical Details:
- The cameras are configured using `media-ctl` and `libcamera-vid` for high FPS capture.
- For stereo vision, I plan to use two GS cameras, synchronized frame capture with calibrated cameras, and post processing with opencv
- Post-processing will align the synchronized frames (since the master video may have more frames) using the PTS timecodes; I will ultimately be focusing on a brief but critical window of 5-7 frames for analysis. (so I'm not concerned with the master taking extra frames initially)
Request for Feedback:
I'm currently outlining the hardware and wiring design for synchronous capture with the dual-camera setup. Given my primary experience lies in software and OpenCV analytics, I'm seeking advice, corrections, or insights on the hardware aspect and the overall design for achieving accurate stereo capture.
Thank you for any guidance or suggestions, especially regarding the setup for synchronous capture and any considerations I might have overlooked.
Statistics: Posted by ddbaron — Fri Feb 09, 2024 7:24 pm