Road Condition Recognition Based On Multi-Source Heterogeneous Data Fusion

With the development of autonomous driving technology,the intelligence and unmanned features of vehicles have become a trend in modern automobiles.However,safety during the driving process remains a critical concern.In order to reduce traffic accidents caused by adverse road conditions and improve vehicle driving safety and stability,this paper proposes a road condition recognition method based on the fusion of multi-source heterogeneous data.The road image and vehicle dynamics data collected by sensor are preprocessed,multi-feature extraction,heterogeneous data fusion and classification respectively.It realizes accurate and efficient identification of road conditions with different weather.The following research is conducted in this regard:(1)Feature extraction of vehicle dynamics data and image dataTo address the issues of low recognition accuracy and poor environmental adaptability in road condition recognition based on single feature identification,a feature matrix of vehicle dynamics data and multiple road image data feature matrices were extracted to improve the accuracy of road condition recognition.Firstly,wavelet thresholding,USM(Unsharp Masking),and histogram equalization were applied for signal and image preprocessing.Secondly,various feature extraction algorithms including power spectral density,color histogram,ULBP(Uniform Local Binary Patterns),and GLCM(Gray Level Co-occurrence Matrix)were used to extract features from both vehicle dynamics data and image data.Lastly,the proposed preprocessing and feature extraction algorithms were subjected to simulation analysis,and experimental verification confirmed that the extracted features can serve as direct and effective descriptors of road surface characteristics.(2)Heterogeneous data fusion and road condition recognitionIn response to the issues of poor timeliness and high operational cost in the process of heterogeneous data fusion,this paper proposes a feature-level fusion-based spatiotemporal fusion strategy for multiple sources of heterogeneous data.Principal component analysis(PCA)is utilized to reduce the dimensionality of the fusion matrix.To address the problem of sensitivity of traditional support vector machines(SVM)to training parameters resulting in a decrease in recognition accuracy,an improved SVM based on the least squares method combined with particle swarm optimization algorithm is employed as the training model for road condition classification recognition.Finally,through simulation and unmanned intelligent vehicle testing experiments,the effectiveness of heterogeneous fusion data in road condition recognition is validated,and the improved SVM is compared with other algorithms,achieving an accuracy of 89.06%.The results demonstrate that the particle swarm optimization-based least squares SVM algorithm can be applied to practical road condition recognition.(3)Software design for road condition recognition systemThe paper presents the design of a road condition recognition demonstration system.The software of this system is developed on the Windows platform and consists of four major interfaces: data feature extraction,heterogeneous data fusion,model establishment,and recognition result display.It integrates multiple functionalities,including data input,feature extraction,matrix dimensionality reduction,model selection,model training,and result export.Based on the software’s output results,it can provide significant technical support for the safe driving of vehicles,making it of practical value.