Transplantation Of Image Recognition Systems To Embedded Platforms

With the rapid development of embedded technology,it is necessary that PC algorithms are transplanted to embedded platforms in intelligent and portable information processing.This thesis is dedicated to transplant the recognition algorithm of banknote serial number and the algorithm of license plate recognition to ARM platform.The main work as follows:1.This paper proposes a method to locate the upper and lower boundaries of the serial number by combining the edge pixels and foreground pixels of binary images in horizontal projection.This method can effectively depress background of serial number images.2.A characters segmentation algorithm is presented.The algorithm uses a sliding square wave to segment the characters of serial number image into single character images.It can also be applied to other regular character segmentation,such as the segmentation of license plate.The experimental results show this segmentation algorithm is robust.3.We introduce a probabilistic modeling method based on Sigmoid function to measure the score of character recognition by SVM.4.We improve the algorithm of license plate recognition based on surveillance video,including proposing a vehicle detection algorithm based on local inter-frame difference,and improving tilt correction algorithm.The improvements increase the computing speed.5.We improve storage structure of LIBSVM model with sparse storage.This makes more than 85%reduction in storage space and the times of multiplication.Also this paper optimizes the algorithm code using floating-point to fixed-point,reducing the use of dynamic memory and other methods to increase the computing speed.