Investigation of MobileNet-Ssd on human follower robot for stand-alone object detection and tracking using Raspberry Pi

Abstract

Human following is a very useful task in the robotics industry. With modern compact-sized robots, there is a demand for further investigated computer-vision solutions that can perform effectively on them. A well-known deep learning model along this line of thought is the MobileNet-Ssd, an object detection model renowned for its resource-constrained usage. Available in popular frameworks like TensorFlow and PyTorch, this model can be of great use in deployments on robotic applications. This research attempts to investigate the MobileNet-Ssd model in order to evaluate its suitability for stand-alone object detection on a Raspberry Pi. To determine the effect of input size on the model, the model’s performance has been investigated with speed in frames-per-second across different input sizes on both CPU and GPU-powered devices. To evaluate the model’s effectiveness in the human following task, a Raspberry Pi-based robot was designed leveraging the tracking-by-detection approach with TensorFlow-Lite. Furthermore, the model’s performance was evaluated using PyTorch while the model’s inputs were adjusted, and the results were compared to those of other state-of-the-art models. The investigation revealed that, despite its modest speeds, the model outperforms other noteworthy models in PyTorch and is an ideal choice when working with Raspberry Pi using TensorFlow-Lite.

Keywords:

• Raspberry Pi
• object detection
• human follower robot
• robot vision
• deep learning
• real-time

Reviewing Editor:

• D. T. Pham, University of Birmingham, United Kingdom

Subjects:

• Artificial Intelligence
• Computer Engineering
• Computer Science (General)

Acknowledgment

The authors would also like to thank Mr. Vishwas G. Kini, Research Scholar, MIT, Manipal, for his assistance in capturing video data for the experiments.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data, metadata and codes used in this research work will be made available upon reasonable request to the corresponding author’s email-id.

Additional information

Funding

This work was supported by the Manipal Academy of Higher Education Dr. T.M.A Pai Research Scholarship under Research Registration No: 200900143-2021.

Notes on contributors

Vidya Kamath

Vidya Kamath is currently pursuing her PhD in the Department of Computer Science and Engineering, Manipal Institute of Technology, Manipal Academy of Higher Education, Manipal, India. She obtained her B.E. degree from Visvesvaraya Technological University, Belagavi, Karnataka, India (2011); and her MTech degree from Kurukshetra University, Haryana, India (2014). She has six years of teaching experience and three years of research experience. She has published papers in international journals and conferences. Her major research interests include computer vision, deep learning and the Internet of Things. Other areas of interest include artificial intelligence, computer graphics, embedded application development and web designing.

Renuka A

Renuka A obtained her BE and MTech degrees from Mysore University and PhD from NITK Surathkal, India. Currently, she is working as a professor in the Department of Computer Science and Engineering at Manipal Institute of Technology, Manipal Academy of Higher Education, Manipal, India. Her areas of interest include cryptography and network security, crypt-analysis, image processing and pattern recognition, machine learning, deep learning, 4G LTE systems and mobile ad-hoc networks. She has published several papers in international journals and conferences. She is also the reviewer for articles submitted to various reputed journals and conferences.