Research And Application Of Vehicle Information Recognition Based On Digital Images

Vehicle information based on the digital image recognition system is an important part of intelligent transportation systems, as a cross-disciplinary research hotspot focus on computer vision, image processing and pattern recognition, which related technology has also been widespread concern. Getting vehicle information accurately can provide effective basis for further control and management of intelligent traffic system and even for combating crime, investigating the fake cars. Based on this background, the paper does systemic research and implements some related technologies of vehicle information recognition, and proposes a vehicle information recognition method combining the vehicle logo and car tail text information, the main research contents and creations are as follows:For vehicle logo information recognition section, this paper proposes a vehicle logo localization and recognition method based on ASIFT (Affine Scale Invariant Feature Transform) algorithm, and realizes the functions of vehicle logo localization and vehicle logo recognition. Vehicle logo localization based on a prior knowledge between the car and the license plate, first, using the license plate localization method for positioning the plate, then according to the positional relationship between the license plate and the vehicle logo to get initial region of the vehicle logo; the initial region using edge detection to get the edge of the image, then use mathematical morphology to expand and corrode the image, and split the logo area and background area to give an accurate vehicle logo localization area. In this paper, ASIFT algorithm is used to extract feature from the logo image, and template matching is used for recognition. Advantages of ASIFT algorithm are invariant in rotation, luminance variation and scale, it also maintains a high stability in changes of the shooting angle, radiation transformation, noise and obscured image.For car tail text information recognition section, this paper proposes a car tail text information localization and recognition method based on SUSAN (Smallest Univalue Segment Assimilating Nucleus) algorithm, and realizes the functions of car tail text information localization and recognition. Car tail text information localization could be divided into two sub-tasks:robust test area localization and accurate test area localization, this paper uses the positional relationship between the text area and the license plate to locate the initial region, then a robust localization area image is captured; edge detection algorithm is used for accurate localization, slips the text area from the background based on the vertical and horizontal projection. The car tail text matching recognition which proposed in this paper creatively implements SIFT(Scale Invariant Feature Transform) to extract features and match templates, searching the corresponding image in the template database, the one which has the most matching points regards as the recognition result, the process is a new application of SIFT.This paper proposes a vehicle information recognition method combining the vehicle logo and car tail text information, which is different from the current domestic and foreign scholars who focus on vehicle shape, size and logo, adds car tail text information recognition to enrich information extracted from the vehicle image, finally, the method presents a relatively complete vehicle information.The article makes full use of the advantages of the algorithm; implements the vehicle information recognition process based on digital images, and the validity and practicability of the method are verified by the experiments.