Follow Me
a project which implements a visual tracking and control system
This “Follow Me” project implements a visual tracking & control system with using a PixKit-1.0 Chassis. Check more details on my github. Meanwhile, a invention patent is authorized in China, check this Google Patent : Link.
I designed and implemented a pure visual tracking system for a customer. The customer expected that this tracking system finally could track three objects(Cars, Pedestrians and UAVs) and follow them once a goal, means SOT(Single Object Tracking). Followings are two video demos.
Requirements and Motivation
Motivation
According to the definition of “Follow Me” functionality, we want our robots(Pixkit-1.0 with cameras) to follow a goal(The user/customer) freely, or we want to add more trailers behind a robot to follow a person who want to use trailers to carry on stuff with hands free.
Hardware Requirements
- Pixkit-1.0 Chassis
- Gimbal Camera(3-Axis Rotation: Roll Pitch Yaw)
- IPC with a GPU card(RTX2070 or better)
Gimbal Camera
The gimbal camear used in this project looks like pictures below.
Software Requirements
- Pixkit-1.0 ROS Driver
- Gimbal Camera control ROS Driver
- Docker
A flow diagram
Detection, Tracking and Visual Control
Future Work
In order to empower Pixkit-1.0 with a visual-based tracking and following funcionality, and mainly focus on tracking a pedestrian freely, we can use aforementioned pure visual tracking system which is the first version. In the next stage, I will develop a second version to cover more categories if needed and optimize selected sensors, I plan to change the gimbal camera to a normal RGB camera(FILR/leopard/Sensing) with a low cost.