Applications Of SVD And Principal Component Analysis In Vehicle Type Recognition

In the Intelligent Transportation Systems, to count and get the information of automobile is the pivotal guidance for the foundation of the vehicle auto-inspect and road auto-tolling system. And the vehicle type recognition technology is important for the automatization of road traffic, the further research of which can enhance the automatization pace of road traffic system and has the gravely actual meaning for promoting the development of Intelligent Transportation Systems. But vehicle type recognition is an unsolved problem, because the background of vehicle image chops and changes and vehicle outer shape is various. Its research attracts many authors; therefore, the vehicle type recognition problem is studied. The main researches and results can be described as follows:(1) Vehicle image pretreatment: For the particularity of the vehicle image of this paper, the image pretreatment includes grey scale and background processing. Grey scale processing is to convert the colorful image to grey scale one, in order to reduce storage space and raise operation efficiency. A background processing method based on phase congruency is put forward. The experiments show the method is resultful.(2) Feature extraction: We extract vehicle feature based on SVD (Singular Value Decomposition) and PCA (Principal Component Analysis) respectively. SVD can reflect the algebra character of the image effectively, and then through SVD operation on vehicle image matrix which includes singular value dimensionality reduction and singular value vector compositor, the feature subspace for recognition is established. PCA is based on statistical theory and principal components of vehicle image can be got through PCA operation on vehicle image vectors. The experiments show that feature extraction based on SVD is more effective than PCA,but PCA has its own advantages in arithmetic complexity and efficiency when the amount of samples increase.(3) Classification method: Two classification methods are brought forward according to the two different feature extraction methods. The minimum distance method is used to match SVD feature extraction and the truncation error method is for PCA. Experiments show the two methods are exact and efficient.