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

  1. Pixkit-1.0 Chassis
  2. Gimbal Camera(3-Axis Rotation: Roll Pitch Yaw)
  3. IPC with a GPU card(RTX2070 or better)

Gimbal Camera

The gimbal camear used in this project looks like pictures below.

Software Requirements

  1. Pixkit-1.0 ROS Driver
  2. Gimbal Camera control ROS Driver
  3. Docker

A flow diagram

Detection, Tracking and Visual Control

Left: YoloX Detection result Middle: Siamfc++ Tracking result Right: Visual Control
Realtime tracking-by-detection and following a pedstrian.

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.