Automatic Label Creation Framework for FMCW Radar Images Using Camera Data

Data acquisition and treatment are key issues for any Deep Learning (DL) technique, especially in computer vision tasks. A big effort must be done the creation of labeled datasets, due to time this task requires its complexity cases where different sensors used. This is case radar imaging applications, data difficult analyze manually. In paper, a semi-automatic framework generate labels range Doppler maps (radar images) proposed. technique based on sensor fusion approach with camera sensors. The proposed scheme operates two steps: first step environment features extraction, which preprocessed filtered remove ghost targets detect clusters, used extract information targets. second step, rule-based system that considers extracted fuses data. By using framework, experimentation performed suggests required label reduced as well possibility human error during labeling task. Our results show can improve final model accuracy regards traditional method, carried out by experts.