基于细粒度分类的汽车家族化计算分析方法

A Computational Method for Car Family Analysis Based on Fine-grained Classification

为了提升汽车主机厂前脸造型的研发效率和自主品牌家族化建设水平,提出基于细粒度分类的汽车家族化计算分析方法。创建了大规模的中国市场汽车前脸图像训练库,利用多级特征表示的机器视觉方法,对汽车前脸造型数据精确分类,最后根据分类结果进行前脸造型的家族化分析。汽车家族化造型分析内容包括品牌造型核心基因,"家族化脸谱"和品牌间前脸造型的相似性分析。基于开发的AutoMorper/iBrandGene软件,试验结果验证了该方法的可行性和有效性。

To promote the research efficiency of car styling design in car factories and facilitate the construction of styling family design for independent brands,a car family styling analysis scheme based on the fine-grained classification was proposed.Firstly,a large-scale car frontal styling database based on Chinese market was created;then,the machine vision method of multi-stage feature representation was adopted to analyze and classify the big data of car frontal styling images;finally,the car family styling was analyzed based on the classification results.The car family styling can be analyzed in three aspects:the kernel brand gene representation of car brands;“The car family face”;styling similarity analysis among car brands.Based on the developed AutoMorpher/iBrandGene,numerical experiments verify the practicability and effectiveness of the proposed methods.