Research On Intelligent Detection System Of Traffic Signs Based On Machine Visio

Traffic sign recognition system is an important research direction in intelligent transportation systems,which collects images or videos through an assisted driving system to provide drivers with traffic sign information,including warning signs,direction signs,and prohibition signs.This system helps to reduce traffic accidents and improve driving safety.Nevertheless,traffic sign detection is confronted with a number of issues such as severe image degradation,false detection,missed detection,and low accuracy,which pose a significant challenge to its efficiency and accuracy.As such,accurate and efficient recognition of traffic signs remains an urgent problem to be addressed.This paper focuses on two critical aspects of traffic sign recognition in natural environmental conditions,namely image restoration and image recognition,and proposes the following research work:(1)To tackle the issue of image blur degradation caused by camera shake and fast vehicle movement,this paper proposes a TV regularization model based on pseudo-inverse fidelity term,coupled with a fast deblurring algorithm.This involves constructing a pseudoinverse matrix to establish the pseudo-inverse fidelity term for image restoration,which solves the issue of difficult solving caused by the traditional observation matrix not being a square matrix.A TV regularization image restoration model based on the pseudo-inverse matrix is then established by combining the noise removal and detail preservation capabilities of TV regularization term.Finally,the objective function is decomposed into four simple subproblems using the idea of split solution,which enables fast solution.Experimental results reveal that the proposed TV regularization model with pseudo-inverse fidelity term has superior noise removal ability and detail preservation characteristics than the traditional TV regularization model.(2)To address the issue of image degradation caused by electronic component noise during transmission,this paper proposes a denoising prior model based on an improved UNet++.This involves the use of skip connections to connect each same-scale feature map in the process of feature extraction and image reconstruction,which enriches the high-level semantic information in the image and strengthens its denoising capability.Furthermore,residual blocks are added after each convolutional layer to protect the integrity of the information.Large-scale noise is used for training to solve the problem that traditional models can only adapt to a single noise level.Experimental results indicate that the proposed model has better noise suppression effect and is universally applicable to large-scale noise without the need for separate training for each noise level.(3)To address the issue of low recognition rate for small traffic signs in images,this paper proposes an improved YOLOv5s-based traffic sign detection model.This involves the use of the SIo U loss function to improve the accuracy of the network by considering the mismatch direction between the real box and the predicted box,which is not considered in the traditional loss function.A small object prediction layer is added to enrich the deep semantic information of the network and improve the accuracy of small object detection.An attention mechanism module is then added to improve the recognition rate of traffic signs by the network.Experimental results demonstrate that the proposed model effectively improves the detection accuracy of small traffic signs and meets the actual detection requirements.