Graph attention network for detecting license plates in crowded street scenes

摘要: Abstract Detecting multiple license plate numbers in crowded street scenes is challenging and requires the attention of researchers. In contrast to existing methods that focus on images are not with vehicles, this work we aim at situations common complex, for example, city environments where numerous vehicles different types like cars, trucks, motorbike etc. may present a single image. such cases, one can expect large variations plates terms quality, backgrounds, various forms occlusion. To address these challenges, explore Adaptive Progressive Scale Expansion based Graph Attention Network (APSEGAT). This approach extracts local information which represents irrespective vehicle because it works pixel level progressive way, identifies dominant include other parts drivers pedestrians, background objects. overcome problem, integrate concepts graph networks scale expansion networks. For evaluating proposed method, use our own dataset, named as AMLPR, contains captured time span, benchmark dataset namely, UFPR-ALPR, provides vehicle, another called, UCSD, cars orientations. Experimental results datasets show method outperforms effective detecting scenes.