| With the fast development of automobile industry in China,the increasing of car brands and models brings great challenge for the efficient design of car styling and the intelligent management in the transportation.For one thing,the traditional car styling analysis and design methods usually involve much human labors and are very costly.Therefore,it is significant to achieve the automatic and efficient analysis and design of car styling,especially when a new vehicle model comes to the market.For the other,the leap of vehicle numbers brings great challenge for the traffic safety.The real-time surveillance of automobiles and the recognition of car brands and models will greatly contribute to the intelligent transportation.To solve the automatic analysis of car frontal styling and the efficient recognition of car brands,a car brand classification method is adopted based on the car frontal styling,and this paper achieves the preliminary research of family property in the car frontal styling and the efficient recognition of car brands.To achieve the automatic and intelligent analysis of car frontal styling,this paper proposes a data-driven approach to the computational styling analysis,which employs the machine learning techniques to discover and analyze styling patterns from the car frontal styling database.In specific,this paper adopts the Car Frontal Styling Database – CFSDB created by the AutoMorpher research group,and applies the PCANet/SVM technique to classify car brands based on the CFSDB and discover the family property of brand styling.This paper also designs a series of visualization methods for the presentation of discover family styling features.The experiments’ results proves that our approach can discover the styling consistency within car brands,styling similarity among car brands,and analyze the brands’ family features.For the video based car brands recognition,this paper proposes to achieve the car brand classification and recognition based on the car frontal styling area.Firstly,the PCANet/SVM technique is applied to train a classifier based on the CFSDB.Secondly,this paper employs the Fast R-CNN to detect the car headlights from the surveillance video,which can be used for the frontal styling area detection.Lastly,the car brand of the detected car styling samples can be recognized based on the trained classifier.In the experiments,the cross-validation accuracy of above 95% within the database and the 90% recognition accuracy in the wild validate the effectiveness of our method.This paper also introduces two software developed based on the PCANet/SVM method and the CFSDB,which are the car frontal styling analysis system – AutoMorpher/iBrandGene and the car brand recognition system – AutoMorpher/iBrandReco. |
