Research On Face Recognition Technology For Traffic Monitoring

Face recognition is a hot research topic in the field of biometric identification, and it plays an important role in security applications. The driver's face image, as an important feature of the vehicle, can provide greater help for the investigation of illegal vehicles. Therefore, it has a broad application prospect and research value for the recognition of driver's face image. The driver's face image captured by the traffic monitoring system is often affected by the varying illumination, such as reflected light of the car glass, sun light, street lamps and so on. In general, there is only one face image for each driver in the system, so it belongs to the problem called face recognition using a single training sample. Therefore, it is focused on illumination processing and face recognition using a single training sample in this paper, which are the two key issues of the recognition of driver's face image. The main research contents of this paper are as follows:(1) Research on illumination invariant feature extraction algorithm for face image. An illumination invariant feature extraction method called Gradientfaces is studied. This method first gets image gradient via performing convolution between the face image and the first derivative of Gauss function, then extracts the illumination invariant features of the face image in the gradient domain. Further, Gradientfaces is improved, and Discrete Gradientfaces is further studied. This method handles the illumination by extracting the more sparse feature in the gradient domain. Experiments show that Discrete Gradientfaces has a better effect.(2) Research on face recognition using a single training sample. Firstly, an image decomposition method that uses QR-decomposition with column pivoting is studied. This method can generate additional virtual samples, so it can transform the single sample into the multi samples. Then, a face recognition method via sparse representation is studied. Further, by combining this method with the general learning framework, a new face recognition method via extended sparse representation for a single training sample is further studied. In this method, the test sample can be represented by a sparse linear combination of the training samples and the intra class variations extracted from the general learning framework, so it can solve the problem of a single training sample. Experiments show that this method has a better recognition effect.(3) Design and implementation of vehicle characteristic database management system. The system requirement analysis, as well as the database design and implementation are performed. Oracle is used as the database management system to store the basic characteristics of the vehicle. MFC is used to write the system frame, and C++is used to write the core modules which include feature extraction and recognition for the driver's face image, set of sign vehicles query, hit and run query and so on.