An Integration of a Neural-Network-Based Computer Vision Model and a 2-DOF Object Tracker Robot

Abstract

This paper presents a system for object tracking that integrates a neural-network-based computer vision model and a 2-DOF robot. The system consists of two main parts: the first part detects objects in the environment and enables the user to select a target for tracking, while the second part establishes a control loop and tracks the selected target using a PID controller. The system uses Yolo neural network for object detection and StrongSort neural network for sorting and assigning IDs to objects. Two methods are proposed for distinguishing the target right before beginning the control loop phase: Landmark-Based Identification (LBI) and Feature-Based Identification. LBI detects a specific and unique sign in the environment to locate the target, while Feature-Based Identification utilizes the StrongSort neural network to extract features of detected objects and compares them to match the desired target. Finally, the tracking procedure begins with the help of StrongSort model and the system calculates the position of the target, generates a control signal to track it even if it’s moving. This system demonstrates the potential of integrating advanced technologies for object tracking applications.