Robust Person Following Under Severe Indoor Illumination Changes for Mobile Robots: Online Color-Based Identification Update

Abstract

Tracking a specific person in environments with non-uniform illumination is a difficult task for mobile robots. Image information such as color is essential to identify a target person. However, the information is not reliable under severe illumination changes unless the system can accommodate these changes over time. In this paper, we propose a robust identifier that has been combined with a deep learning technique to accommodate varying illumination in the ambient lighting of a scene. Moreover, an enhanced online update strategy for the person identification model is used to deal with the challenge of drifting the target person's appearance changes during tracking. Using the proposed method, the system achieves a successfully tracked rate above 90% on real-world video sequences in which variations in illumination are dominant. We confirmed the effectiveness of the proposed method through target-following experiments using five different clothing colors in a real indoor environment where the lighting conditions change extremely.