AUTOMATED REAL-TIME ROAD SIGN DETECTION FOR GPS ERROR CORRECTION BASED ON AN IMAGE SENSOR

摘要

In these days, development of related vehicle, vision and geospatial technology expedite the development of autonomous vehicles. The key to developing the autonomous vehicle is knowing exactly where the car is located. For this, GPS is used but there are problems such as GPS signal losses in urban area or underground. To complement this, we propose a method for automated real-time road sign recognition for GPS error correction based on an image sensor. For this purpose, we need to know exactly where road signs are in the images taken by an image sensor and to which road sign templates in road sign database they are corresponding. Due to the characteristics of autonomous vehicles, we must perform road sign recognition in real-time and automatically. In the proposed method, there are three steps : In the first step, we apply two operations to the road driving image. The one is to segment the image. So that candidate road sign area is located within the segment. The other is to convert RGB scale image to HSV scale. This is for searching specific color located region without influence of illumination. In the case of HSV image, we make only the green or blue color area to be displayed. Through the first step, we can extract green and blue color region of driving image. In the second step, we find rectangle feagures from the first step. This step can reduce the number of ROI candidates. In the third step, we match between ROIs of driving image and road sign template within database. Through this process, we can estimate relationships between driving image and road sign template. For experiments, we used images obtained from actual road driving. Template images were cut out from one image of the images. After the three steps, we could confirm the exact position of the template image in the road driving image. These whole steps were processed automately and applicable to real-time applications. It is expected that the position of the sensor mounted on the vehicle will be accurately estimated.