Occupant Attributes Recognition for Thermal Comfort in Passenger Car Cabin

Occupant attribute recognition is a key issue in the evaluation of thermal comfort in passenger car cabins. Different gender, ages, and dress make passengers have different perceptions of the temperature in the car. We developed an occupant attributes recognition system that utilizes the image taken in the cabin to recognize the gender, age, and clothing type of occupants. The system incorporates open-source object detection models and a multitask learning network, which were trained on various publicly available datasets. To address the issue of low recognition accuracy caused by the lack of in-cabin scene images in existing public clothing datasets, we collected 31 images of occupants wearing different clothes. Through the analysis of experimental results, the proposed passenger attribute recognition system demonstrates high accuracy and fast recognition speed in real-world driving scenarios.