Analysis And Generation Of Vehicle Front Face Styling Based On Deep Learning

With the increasing consumer demand for vehicle styling aesthetics and the increasing trend of vehicle styling diversification,improving the automation and intelligence in the vehicle styling design process has become a key issue that needs to be resolved.Through the analysis of vehicle brand styling,styling designers can better grasp the design ideas and development trends,and carry out innovative styling design on the basis of ensuring and strengthening the family characteristics of the brand.At the same time,the use of automatic analysis results for the automatic generation of front face styling can shorten the cycle of styling conceptual design,allowing the OEM to remain active in the highly competitive market.In view of the above vehicle styling design problems,this paper proposes a deep learning-based vehicle front face styling analysis and generation method,which achieves vehicle front face intrinsic styling feature extraction,component family analysis and automatic generation of styling images.Aiming at the problem of automatic extraction of intrinsic styling features of the vehicle front face,this paper proposes the detection and elimination method of sensitive areas,realizes the accurate recognition of the vehicle brand based on the convolutional neural network classifier and achieves the extraction and visualization of the intrinsic styling features of vehicle front face.A data-oriented augmentation method for the region of interest in the vehicle front face for deep learning is proposed in this paper to enhance the extraction of intrinsic styling features.In addition,this paper proposes a large-scale vehicle front face database FVD22-Cars,which contains 22,316 pieces of vehicle front face data of 22 domestic mainstream vehicle brands.The database contains original dataset,occlusion-invariant dataset,styling feature dataset and challenging testing dataset.Numerical experiments show that this method can effectively extract the vehicle front face styling features and significantly improve the accuracy and robustness of the brand recognition model.Aiming at the problem of component family analysis of vehicle front face,in order to reduce the interference of subjective factors as much as possible,this paper achieves data preprocessing and component extraction based on the front face wireframe data set,and uses K-means algorithm to cluster outlines,windows,headlights,fog lights,and grilles.Then this paper visualizes the clustering centers of various components and analyzes the components distribution of various brands.Benefiting from data driving and automation,the method achieves the division and analysis of the component-level styling of the vehicle front face.Combining the FVD22-Cars database and the clustering results of the vehicle front face components,this paper calibrates the brands,headlights,fog lights,and grilles categories of the data and establishes the VSA22 vehicle front face styling attribute data set.Aiming at the problem of oriented generation of vehicle front face styling images,this paper connects ACGAN network and MS-CNN network to automatically generate vehicle front face styling images through model splitting method.This paper provides high and low resolution outputs to meet different hardware conditions and the actual needs of users.The experimental results show that the method can effectively achieve the oriented generation of the vehicle front face styling image,and initially achieve the automation and intelligence of the conceptual styling design.