Skip to main content

MIPI Camera Sample

Introduction​

The MIPI camera sample is a Python API example under /app/pydev_demo/03_mipi_camera_sample that shows how to use the on-board MIPI camera for real-time object detection. It runs an FCOS model on the live stream and draws results on HDMI while printing boxes and FPS.

Demo​

output-img

Hardware setup​

Connections​

  1. One RDK board
  2. Official compatible MIPI camera
  3. HDMI cable to a display
  4. Power and Ethernet

connect-img

Quick start​

Code location on device​

The sample file is /app/pydev_demo/03_mipi_camera_sample/mipi_camera.py.

Build and run​

Python samples do not require compilation; run directly.

Sample output​

After start, the program initializes the MIPI camera and HDMI, runs real-time detection, draws on HDMI, and prints detections and FPS in the console.

Details​

Command-line options​

No arguments are required; the on-board MIPI camera is detected automatically.

Software architecture​

  1. Load the FCOS model
  2. Initialize the MIPI camera with srcampy.Camera()
  3. Initialize HDMI display
  4. Bind camera output to display
  5. Real-time loop: grab frame → infer → post-process → draw → print FPS

software_arch

API flow​

API_Flow

FAQ​

Q: Camera initialization fails.
A: Check the MIPI connection and drivers; reboot if needed.

Q: No or bad HDMI output.
A: Check HDMI and stop the desktop manager if required (systemctl stop lightdm).

Q: How to change the score threshold?
A: Edit fcos_postprocess_info.score_threshold (e.g. 0.5).

Q: How to change display resolution?
A: Adjust sensor_width / sensor_height if the monitor supports it.

Q: Low FPS.
A: Try a lighter model or lower capture resolution.

Q: How to save result images?
A: Add saving logic such as cv2.imwrite().

Q: How to add classes?
A: Update get_classes() and use a model trained for those classes.

Last updated on