Computer Vision, 3D tracking of a person in lock mode (no yaw)

pgminin · Oct 6, 2019

Tello SDK
Object Detection: Yolov3 trained on COCO
Control loop @2fps
3D PID tracking, set point distance: 750-800 cm

guy250! · Oct 8, 2019

great work

TelloPilot01201956 · Jan 20, 2021

Very nice work! I am thinking of developing an app for tello to track me on my mountain bike. Do you think that is possible?

pgminin · Jan 20, 2021

TelloPilot01201956 said:
Very nice work! I am thinking of developing an app for tello to track me on my mountain bike. Do you think that is possible?

In sdk mode, the easy way to control our Tello, max speed is only about 2 m/s, so sadly it's too slow for a bike!
Here you can see my demo:

TelloPilot01201956 · Jan 20, 2021

So thats about 7km per hr. Thats rather slow. I wanted something like 15 km per hr at least.
Is recognizing a biker a problem too?

pgminin · Jan 20, 2021

TelloPilot01201956 said:
So thats about 7km per hr. Thats rather slow. I wanted something like 15 km per hr at least.
Is recognizing a biker a problem too?

Using low level protocol (not just sdk) Tello can go faster, .
You can test computer vision capabilities using my free Android App:

Tello Vision 1D - Apps on Google Play

Computer Vision and 1D Follow Me for DJI Ryze Tello drone (SDK mode)

play.google.com

volate!lo · Jan 20, 2021

Tello can do only 2-2.5 m/s even with low level protocol if used with computer vision.
Fast mode is possible for the low level protocol but then there is no video stabilization so unsuitable for computer vision.

multipass0311 · Mar 14, 2021

volate!lo said:
Tello can do only 2-2.5 m/s even with low level protocol if used with computer vision.
Fast mode is possible for the low level protocol but then there is no video stabilization so unsuitable for computer vision.

In fast mode, use digital stabilization as a preprocessing step to your computer vision pipeline: