Mobile robots operating in seniors' homes can serve as social companions and assist with daily tasks, thus enhancing the seniors' quality of life. In order for robots to assist seniors, it is crucial that they are equipped with sets of social and interactive skills to enable them to have natural and personalized interactions. Personalized interactions, such as using patients' proper names or remembering personal preferences, is necessary to establish strong social relationships, and is a key factor to improve trust in human-robot interaction. A prerequisite for robots to achieve personalized interactions, however, is the ability to automatically recognize and re-identify people around them. Existing person re-identification systems for mobile robots are highly restricted in terms of where robots can operate, and do not stimulate natural and personalized interactions because they need preliminary knowledge about the robot's users, rely on facial cues, or use data collected from external sensors. This thesis introduces two lightweight Siamese convolutional neural networks, LuNet Light and LuNet Lightest, designed for the problem of person re-identification in a robotic setting without relying on the aforementioned restrictions. Despite being significantly more lightweight than other person re-identification systems, LuNet Lightest achieves near state-of-the-art results on the MARS dataset evaluation protocols. This thesis additionally presents a set of evaluation measures tailored to evaluate re-identification systems for robots operating in various environments. When simulating crowded environments, LuNet Lightest reaches 92.4% balanced accuracy on the proposed evaluation protocol. As a result of the lightweight architecture, LuNet Lightest achieves real-time frame-rates of 71.6 frames per second when using a GPU, 33.9 frames per second when using a CPU without GPU, and 15.7 frames per second when using only one core of the same CPU, rendering the proposed system highly suitable for low-cost, hardware-constrained robots. The proposed person re-identification system will enable assistive mobile robots to robustly and accurately identify their users, and is a preliminary step to improve trust and attain natural and personalized interaction between robots and patients.