Multi-source domain fusion cross-domain pedestrian recognition based on high-quality intermediate domains

Pedestrian detection has received considerable attention over the last few years because technology can be combined with pedestrian tracking and re-identification in areas such as vehicle-assisted driving intelligent video surveillance. However, existing techniques have achieved excellent results. Problems domain gaps lead to poor generalization performance, thus limiting its application practical value. This paper proposes a High-Quality Integration Domain framework for recognition. First, source domains are produced super-resolution training data. The HCycleGAN model uses algorithms generative generate high-quality intermediate domains. Second, multi-source fusion scheme based on NPIQE module is proposed improve generated framework’s quality reduce dataset’s overfitting. It fuses images by three aspects: similarity, blurriness unsupervised image score values. Finally, we use an anchor-free Center Scale Prediction detection. experimental dataset contains two common datasets, Caltech CityPersons. Cross-domain results show that cross-domain from CityPersons 6% 5%. of achieves almost same accuracy original domain. In conclusion, this effective provide ideas inspiration future applications.