Research And Application Of Traffic Sign Detection And Recognition Technology

As a significant component part of intelligent transportation system,traffic sign detection and recognition technology has played an important role in automatic traffic sign maintenance,assist driving or automatic driving.In view of the problems in existing traffic sign detection and recognition technologies,this paper studies the following aspects:Image dehazing as pretreatment.Extreme hazy weather has a serious impact on the imaging quality of computer vision systems.To solve this problem,this paper proposes a fast dehazing algorithm under driving scenario based on dark channel prior and guided filter.The improved algorithm can still maintain real-time performance on medium-size images and provides reliable input image quality for detection and recognition algorithms.Traffic sign detection algorithm.In view of the poor adaptability and high complexity of the traditional color-based or shape-based detection methods,a traffic sign detection algorithm based on multi-scale input and fully convolutional network combined with non-maximum suppression is proposed in this paper.The algorithm achieved a good detection effect.Traffic sign recognition algorithm.Aiming at the problem of poor adaptability,high complexity and bad scalability in the traditional recoginition methods,this paper designs a traffic sign recognition algorithm based on convolutional neural network,combining data enhancement with Dropout to increase the effectiveness of the model.The algorithm achieves a recognition accuracy of 98.09%on the test dataset.Prototype system design and algorithm performance analysis.In view of the practical application of traffic sign maintenance,a prototype system of traffic sign automation maintenance is designed and its core functions are realized.Simultaneously,three indexes of detection recall,recognition accuracy and system,accuracy are designed to measure the detection algorithm,the recognition algorithm and the overall performance of the system.Through the performance analysis on the prototype system,the dehazing algorithm,the detection algorithm and the recognition algorithm designed in this paper are highly effective in the face of various complex scenes.