At present, most LPRS (License Plate Recognition System) is designed to capture andrecognize vehicles in a single lane. Because it is difficult to capture vehicle images in multilanesat the same time. In order to recognize vehicle license plates in multi-lanes, it isessential to capture high-resolution images. In addition, t should be little blooming and smearand good infrared sensitivity are required for capturing images under night illumination. TheCMOS image sensor is suitable to be used in an outdoor environment such a LPRS in that ithas a high-speed image capturing, little blooming, no smear and wide dynamic range.However, the high resolution CMOS sensor has a problem to capture images of an objectmoving at high speed, using a rolling shutter. In this paper, we propose a method ofcontrolling cheap high-resolution CMOS image sensors in order to obtain vehicle images inmulti-lanes. First, the image sensor is installed with the 90-degree rotation. Images are caughtat a high speed while keeping a desired resolution by carrying out down-sampling in theentire area (global scene, sub-sampled image) and re-sampling of the region of interest(vehicle scene, ROI image) with a time difference in acquiring sensor data. We calculate thedriving speed of a vehicle in the sub-sampled image and position-prediction of the licenseplate of the vehicle for the ROI image, using the calculated speed. And then a vehicle licenseplate is recognized by corrected ROI image.
展开▼