Object tracking system

PX4 Autopilot
1

Hello
I am working on my university graduation project, which is as follows:
I am building a plane with a px4, I also connected it to a jetson nano kit
And I connected it to a camera with an object tracking system
my question is:
Can I control the drone through the (computer vision) (object tracking) in the camera?
Or should I start programming a new object tracking?
thanks for the help

2

Hi Dark, do you have any follow up to this thread?

3

Just starting, and I want help to know what the main things I have to focus on

1 Like
4

Yes you can use Pixhawk FC and Nano for this task!
May I know more about the model you are using and whether it is just the tracking as your end objective by the camera feed or something else?

5

Hi everyone,

I’m working on a UAV project where I want to detect, track, and follow a moving object without using GPS (fully GPS-denied environment). I would really appreciate guidance on both camera selection and localization strategy.

:puzzle_piece: Current System Architecture

  • Flight Controller: Pixhawk running PX4

  • Companion Computer: Considering

    • NVIDIA Jetson Nano (for onboard AI)

    • Raspberry Pi Compute Module (lighter but less powerful)

:camera: Camera Options I’m Considering

Option 1: Global Shutter Monocular Camera

  • Pros: No motion distortion, simpler pipeline

  • Cons: No direct depth information

Option 2: Luxonis OAK-D Lite (Stereo + RGB)

  • Has stereo depth + onboard AI acceleration

  • Includes global shutter mono sensors for depth

  • Pros: Depth estimation + object detection on-device

  • Cons: More complex integration, potential latency

:bullseye: Goal

  • Detect and track a target (person/vehicle/etc.)

  • Estimate relative position

  • Control the drone to follow the target autonomously

  • Operate fully without GPS

:red_exclamation_mark: Main Challenge: Localization in GPS-Denied Environment

This is where I’m struggling.

If I use:

  • Monocular camera: I only get 2D tracking (no scale/depth)

  • Stereo camera (OAK-D Lite): I get depth, but I’m unsure how reliable it is during motion and at longer distances

:red_question_mark: Questions

  1. Camera Choice:
    Would you recommend a global shutter monocular camera + VIO
    or a stereo system like OAK-D Lite for this use case?

  2. Localization Approach:
    In a GPS-denied setup, what is the most robust way to estimate position for object following?

    • Visual Odometry (VIO / SLAM)?

    • Depth-based relative tracking?

    • Sensor fusion (IMU + vision)?

  3. Tracking + Control:
    What is the best approach to convert:

    • detected object position (image space or 3D)
      → into stable flight control commands?

  4. Practical Experience:
    Has anyone built a similar system?
    Any issues with:

    • latency

    • drift

    • depth noise

    • fast-moving targets

:light_bulb: Additional Notes

  • Real-time performance is important

  • The drone will operate in dynamic environments

  • I can design the rest of the hardware (power, mounting, etc.)