In recent years, with the development of road traffic safety improvement system and intelligent transportation system, appearance modified vehicles have attracted much attention, including the modification of appearance attributes such as vehicle type, color, and license plate. At the same time, vehicle attribute recognition and product quantization retrieval algorithms have also been widely used in the field of intelligent transportation. Vehicle attributes include the vehicle type, color, license plate, and so on. Product Quantization (PQ) retrieval algorithm generally extracts the feature of the whole image at first, and then performs segmentation, clustering, quantization coding and query on all vectors.In the scene of modified car retrieval, the effect of overall feature extraction of the picture is easily affected by the complex background of the picture. And the k-means clustering algorithm used by PQ is sensitive to noise, outliers and the choice of initial center point, which has a great impact on the clustering results. This will make the retrieval accuracy decrease. Aiming at the above problems, this paper proposes a modified vehicle recognition method based on attribute fusion and fuzzy product quantization. The vehicle and license plate are intercepted by vehicle and license plate detection, and then the three kinds of features are extracted in parallel and the attribute fusion and splicing method is used to obtain the comprehensive features that contain not only high-level semantic information of the license plate, but also low-level vehicle type and color attributes. In this way, the interference of background noise is eliminated and the response strength of three features that are easy to be modified is enhanced. The sensitivity to noise and outliers is reduced by introducing membership degree into fuzzy clustering. Fuzzy clustering uses random initialization to select the initial membership, which reduces the dependence on the initial point. In addition, the proposed method can meet the needs of displaying multiple suspicious vehicles similar to the target modified vehicle when the input vehicle information is fuzzy.In this paper, the above two improved methods are combined, and the performance test is carried out on the VOC 2012 datasets through experiments. In the Top30 retrieval results, the average retrieval accuracy of the proposed method is 5% higher than that of PQ.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    Modified Vehicle Recognition Method Based on Attribute Fusion and Fuzzy Product Quantization


    Contributors:
    Peng, Kun (author) / Hou, Qun (author) / Liu, Xin (author)


    Publication date :

    2023-10-28


    Size :

    922948 byte




    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    English



    Improved vehicle identification method based on attribute fusion and fuzzy product quantization

    YANG ZHIXIANG / PENG KUN / YANG XIAOTAO et al. | European Patent Office | 2023

    Free access

    Vehicle Attribute Recognition Algorithm Based on Multi-task Learning

    Sun, Jingying / Jia, Chengzhe / Shi, Zhiguo | IEEE | 2019


    Integration Research of Vehicle Attribute Recognition Algorithm Based on YB-CNN

    Li, Jianbin / Zhang, Yang / Fang, Rui et al. | Springer Verlag | 2025


    On-line handwritten Chinese character recognition via a fuzzy attribute representation

    Chen, J.-W. / Lee, S.-Y. | British Library Online Contents | 1994


    Pedestrian Attribute Distillation Fusion Model

    Ding, Huilin / Gong, Yuzhou / Han, Shoudong et al. | Springer Verlag | 2025