The Design Of Tourists Intelligent Guide System Based On ARM

The counting of tourist number and automatically guide through video is a new application trend of intelligent video supervision. In recent years, the emerge of large-scale exhibition provides an opportunity for Intelligent Guide System (IGS), especially in Shanghai World Expo. How to combine the image processing technology with embedded system to identify tourist number automatically is a new issue. The IGS can improve the efficiency of visits and ensure the tourists safety. It has a high commercial value and potential. The IGS based on ARM was designed. The entire system includes two parts: hardware structure and recognition algorithms.EmbedSky's ARM9 board TQ2440 was as the development platform, and CMOS image sensors OV9650 was adopted to capture the scene. The camera driver was designed and implemented in Linux environment. The sampled frame was generated into BMP file.The original BMP image was processed through ladder low-pass filter, to remove the Gaussian noise. To reduce calculating loads, the Signal to Noise Ratio (SNR) was introduced to judge whether the image need to de-noising. For the motion blur resulting from relative motion, projection restoration method was used to restore the image. The background subtraction method was used to extracted foreground objects. A background adaptive method was proposed to updates the scene background, which can reduced the effect of light change on detecting target.The Canny edge detection algorithm was used to extract contours. Based on image morphology processing theory, through repeatedly erosion and dilation, filling gaps in closed areas, to eliminate misjudgment caused by person shelter and adhesion, a relatively complete closed contour was obtained. A new moment invariants , contour moment invariants was constructed based on the Hu moment invariants, which with the character of shift, rotation and scale invariance, independent of the image gray-level. By analyzing the template image contour moment invariants, setting allowable error threshold, matching the sample image contour, the number of tourists was determined. The automatic guidance can accomplish combing hall capacity and alarm settings.Recognition algorithms were written in Visual C++ environment using OpenCV computer vision library, Makefile was written for port documents to Linux system after debug was completed.The experiment results showed that the accuracy of recognition was high in the environment of enough light and less complicated background. When the light attenuate or the background similar with the foreground, the accuracy come to decline.