Real-Time Person Re-Identification to Improve Human-Robot Interaction

Abstract

Mobile robots deployed in the homes of disabled people and older adults can play a vital role as companions and aid in activities of daily living, enhancing people’s quality of life. For these robots to effectively assist, it can be helpful if they possess social and interactive skills to facilitate natural and personalized interactions. Personalization, including using proper names and remembering preferences, is crucial for building strong social connections and fostering trust in human-robot interaction. However, achieving personalized interactions necessitates the robot’s ability to autonomously recognize and re-identify individuals without relying on specific constraints or prior knowledge. This work introduces two lightweight Dual convolutional neural networks, LuNet Light and LuNet Lightest, designed for person re-identification in a robotic context without the limitations of existing systems. We introduce a lightweight architecture, LuNet Lightest which achieves near state-of-the-art performance on the MARS dataset. It is well-suited for real-time applications on low-cost, hardware-constrained robots. This person-identification system enables assistive mobile robots to accurately identify users, fostering trust and facilitating natural, personalized interactions.