A Dynamic Target Tracking System with DepthAI and The OpenCV AI Kit (OAK-1)
Here’s a quick demo:
A Target Tracking System with MobileNet-SSD, OpenCV AI Kit, Pi 4 & A PT Bracket
How the systems works:
The OAK-1 is used to
- capture live video for the MobileNet to do object detection, and to calculate the offset between the center of the object and the center of the frame.
- convert the offset to Panning/Tilting angles with PID (Proportional-Integral-Derivative) control.
- report the final results back to the Raspberry Pi.
The Raspberry Pi 4 is used to
- power the OAK-1 and the bracket.
- control the OAK-1 to start the camera, run the model and do all the calculation work.
- navigate the Pan-Tilt Bracket to track the object and keep it at the center of the view.
The Bracket is used to
- move the OAK-1 so it follows the object.
Please also note that OAK-1’s onboard camera (IMX378) features an autofocus lens, so it can keep the object in focus while moving along with it. The bracket draws power through the Raspberry Pi’s GPIO pins, it can also be powered externally. Even though object detection is done with MobileNet-SSD, which is a pre-trained model to detect as most as 21 objects, you can use other models to detect and track more things.
Real-World Applications with Target Tracking Cameras
Target Locking Systems
This is widely incorporated as a smart flight feature in these “follow-me drones”, where when a “follow-me mode” is enabled, the quadcopters will self-navigate and follow a given target autonomously.
Follower Robots for Logistics & Consumer Markets
Target recognition and tracking is also widely adopted in supply chains, where warehousing robots are made to improve inventory management efficiency, reduce human errors, and save logistical costs. There are also many humanoids and robotic toys that use target tracking to create enhanced interactions.
Smart video conferencing is also another great use case for target tracking cameras, where when you are having a remote team meeting, the webcam automatically follows the host and switches to another presenters whoever starts talking.
Combining Motion Control with Embedded Machine Vision
Embedded hardware equipped with machine vision algorithms may potentially be the best solution to cut down a huge portion of R&D costs in many motion control applications, and by lowering the manufacturing expenses from the start, this will benefit consumers in both the industrial and retail markets.