Lane Detection And Traffic Sign Recognition Based On Machine Vision

Among the road information,there are the lane lines and signs information.Effective access to the right lane and road sign information will play an important role in driverless vehicles' decisions.Improving the accuracy and real-time performance of lane detection and road sign recognition algorithm is the key to the practical application process.In face of the increasing number of cars and high traffic accidents,it is of great significance to improve the safety of automobile driving.The detection method of lanes in road images is studied.Firstly,the image is pre-processed and the lane edge is effectively extracted by the gradient algorithm.Then,a lane detection method based on improved Hough transform and least squares algorithms is proposed.Finally,the method is tested with actual traffic images in Guangzhou.The experimental results show that the lane detection method applies to a variety of road conditions with strong robustness and good real-time performanceConvolution neural network has good performance for target detection and classification,and the structure parameters of the CNN directly affects its ability of classification.To explore the optimal SSD structure for sign recognition,we investigated how the structural parameters,such as number of iterations,size of input layer and batch size,influenced the recognition results.At last,the SSD was constructed for road sign recognition,which ensured the network had not only the good performance in classification,but also a high efficiency.A road sign recognition method based on SSD network is proposed for the requirement of robustness and real-time in road sign recognition.Through data acquisition,the image data of different weather and light conditions are obtained.Expand the amount of samples,and label the image data with annotation software manually,to obtain the road sign dataset.Analyze the influence of different network parameter selection through experiments,and obtain a suitable parameter combination.Finally,verify the effectiveness of SSD model with experiments.Aiming at the shortcomings of the original model,an improved R-SSD model is proposed,test shows that the improved model improves the model identification precisions.