Research On Prohibition Sign Detection And Recognition Based On Computer Vision And GPU Acceleration

The continuous development of society and economy has made car a common vehicle of ordinary Chinese family,which the increasing number of cars and frequent happening of traffic accidents pose a great threat to people’s life and property 。Therefore,the driver assistance system is of crucial importance to make improvements in driving safety via helping control the vehicle during the emergency,which has significant social values and research values under the existing road environment.Prohibition signs are the most stringent regulations in driving behavior limits,so the detection and recognition system of traffic prohibition sign,an important component of the driver assistance system,is receiving more and more attention.However,the real complex environment,traffic prohibition sign situated,made many difficulties for us such as traffic prohibition sign influenced by illumination change,weather condition,partial occlusion and similar distracters.At the same time,the detection system of traffic prohibition sign requires real-time and portability.Because of the high requirement and many difficulties,the application of detection and recognition system of traffic prohibition sign has just made a few steps.The main research content of this paper is the detection and recognition system of traffic prohibition sign.Starting from the research on the traffic prohibition sign’s color and shape,the paper first gives an analysis on its characteristic and conducts a research on its localization algorithm in real environment,then by deep learning thought delivers an analysis on its classification task and achieves its classification task through convolutional neural network.With the objective requirement of real-time and portability,the paper here conducts a research on parallel realization and embedded implementation of image processing algorithm to improve practicability.The main researches of the paper are as follows:1.Researches on color image enhancement algorithm and color segmentation algorithm.If original image are applied in image segmentation,good effect cannot achieve.In order to solve this problem,the image spatial domain enhancement algorithm and frequency domain enhancement algorithm are studied in this paper as well as the RGB color model and HSI color model,and their corresponding segmentation effect.2.Researches on shape detection algorithm.First calculating the convex hulland the convex defect of the prohibition sign,then calculating the area,length,height and length width ratio of the convex defect.Excluded the false area and segmented the connected mark base on above properties.3.Researches on classification task of traffic prohibition sign and on deep learning theory.Through researches,design a convolutional neural network which is suitable for classification task implied within the paper to achieve successful classification of traffic prohibition sign.The final experiment shows that accuracy of classification can be achieved at 95%.4.Researches on GPU parallel computation and parallel image process algorithm.The paper,gives an analysis on algorithm parallelism for processing the traffic prohibition sign,applying NVIDIA’s embedded TK1 platform to design and implement some algorithms which are suitable for parallel.Base on this device,this article implement the image format conversion,image enhancement,image color segmentation algorithm and convolutional neural network,the parallel mode can achieve 2~9 times faster than serial mode.