数据驱动的车身概念模型自动化三维建模

Data-driven Automatic Surface Modeling for Car Body Structure Concept Design

由于汽车造型设计受复杂因素制约而难以提高其设计效率,所以提出一种基于汽车侧视图的由数据驱动的车身概念模型自动化建模法。对输入侧视图或手绘图进行预处理,提出一种组合车型分类法进行车型识别,再基于二值特征回归法进行车身关键点的自动检测。采用数据驱动的可变体建模法实现自动化二维特征线和三维特征线(网)建模,然后由生成的参数曲线网络进行自动化曲面建模。为实现数据驱动的自动化识别、检测和建模,构建了大规模的汽车侧视图像标注库及其对应的二维特征线训练库,以及二维和三维汽车特征线可变体模型库。大量数值试验表明,该建模流程可自动、高效地生成较高质量的车身二维和三维概念模型。

It is difficult to improve the design efficiency of car body styling due to many complex factors. This paper presents a data-driven automatic surface modeling method for car body structure concept design based on vehicle side view drawings. First of all, we presented a classification method for vehicle type recognition based on the pretreated vehicle side-view drawings or drawings using manual drafting. Then, the landmarks in the vehicle side view images were localized via regressing local binary features. After that, 2D and 3D curves of car body could be reconstructed automatically based on the data-driven morphing modeling method. Finally, the 3D car body surface could be generated through 3D curves. In order to achieve vehicle classification, landmarks localization and 3D curve reconstruction, the database including the vehicle side view images with annotations and 2D and 3D curves of car body was established. A large number of numerical experiments show that this modeling process can generate 2D and 3D concept car body with high quality automatically and efficiently.