In crowded scenes, it is challenging to extract the appearance features of pedestrians due to occlusion issues. To enhance the semantic features of pedestrians, we propose the Pedestrian Attribute Distillation Fusion Model (PADFM). In the absence of attribute labels in pedestrian detection datasets, PADFM acquires the capability to predict pedestrian attributes through knowledge distillation. It integrates pedestrian attribute information into the detection process, thereby increasing the basis for pedestrian discrimination and improving the detection performance of pedestrians in occluded scenes. Furthermore, since the PADFM can simultaneously output detection results and their attribute information, it provides richer information for related applications in the field of pedestrian detection.
Pedestrian Attribute Distillation Fusion Model
International Conference on Informatics Engineering and Information Science ; 2024 ; Shenzhen, China May 17, 2024 - May 19, 2024
05.02.2025
9 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Machine-learned model training for pedestrian attribute and gesture detection
Europäisches Patentamt | 2023
|Machine-learned model training for pedestrian attribute and gesture detection
Europäisches Patentamt | 2024
|Multi-label convolutional neural network based pedestrian attribute classification
British Library Online Contents | 2017
|Multi-label convolutional neural network based pedestrian attribute classification
British Library Online Contents | 2017
|