Marker Based Human Following Robot Using Deep Learning Object Detection for Assistive Load Transportation

The next generation human following robot is the answer to a solution of autonomy in the field of assistive robots in healthcare, logistics, and the public service sector, but current systems suffer from limitations in both computational efficiency and the stability of tracking in a crowd setting. This paper presents a visual marker-based tracking system of an autonomous load carrying mobile robot that uses a custom trained YOLOv8 deep neural network to identify a unique red circular marker on the user, allowing it to perform effective human following behavior with real time performance at resource constrained embedded hardware. In contrast to the traditional full body human detection methods, the suggested marker based system allows for greatly reducing the high calculation burden and removing target confusion in the multi person scene. The system combines computer vision and multi sensor fusion, incorporating infrared sensors, to be able to avoid dynamic obstacles and navigate safely. The robot uses a Raspberry Pi 5 platform with a 720p web camera and can accurately detect markers with a precision of 98.8 %, a Recall of 92.2 %, and a mean Average Precision (mAP) of 97.1%, and can also perform real time inference at 28-32 milliseconds per frame. The lightweight and aluminum body allows it to carry loads of up to 15 kg and have stable autonomous mobility. Extensive experimental testing in diverse indoor environments demonstrates that our system is more consistent in tracking than old conventional human detection techniques, especially when dealing with full crowds, as the marker based system can ensure no false positives. The vertical-line-based tracking algorithm ensures smooth path following with dynamic maintenance of distance, and the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$A$</tex> path Planning algorithm provides efficient obstacle avoidance in real time. The study provides an efficient and cost effective solution to assistive robotics by showing the empirical feasibility of marker based visual servoing of human following in dynamic real life settings.