Human-following robot using deep-learning techniques

Abstract

This work proposes a human-following robot based on deep learning techniques. The system utilizes a deep neural network to detect and track a target in real time, using an onboard camera coupled with an autonomous navigation module for safe operation. Key challenges such as handling occlusions, varying lighting, and real-time processing are addressed. The anticipated result is a robust system applicable to personal assistance, security, and healthcare. The proposed methodology integrates real-time object detection using the YOLOv4 deep learning model with a histogram-based identity lock mechanism for consistent person tracking. The integrated camera captures live video, which is processed locally to detect and follow a human target. Motion commands are computed based on the position and size of the detected bounding box and sent to TurtleBot2 using the Robot Operating System. In experimental tests, the robot maintained an average tracking accuracy of 94.6% with a real-time processing speed of 12-15 fps and a command response delay of 0.3 seconds. These results demonstrate the system’s ability to reliably follow a human target under indoor conditions without the use of additional sensors.