Target Detection Of Traffic Signs Based On SSD Algorithm

With the rapid development of modern science and technology,informatization and intelligence have gradually entered the commercial stage,and technologies related to intelligent transportation systems have emerged.The detection of traffic signs is one of the important research issues of intelligent transportation systems.Aiming at the problem that traditional detection methods have low detection accuracy and it is difficult to accurately detect targets in complex realistic scenes,this paper proposes a traffic indication based on SSD algorithm.The method of brand target detection,and in view of the problem that the SSD model discards low-level feature information,which results in poor detection of small targets and cannot be used well in the implementation scenario.The SSD model based on the feature pyramid FPN is proposed.Algorithmic traffic sign target detection method,combined with the idea of FPN feature pyramid network model to improve the feature representation ability of SSD model.The use of the lightweight module CBAM will not increase the amount of calculation for the convolutional neural network CNN.This module can be inserted into the common convolutional neural network CNN for end-to-end training,which can significantly improve the advantage of target detection accuracy.The module is inserted into the SSD model,and a traffic sign target detection method based on the SSD algorithm of CBAM is proposed.Finally,the proposed method is used to detect the traffic sign.The experimental results show that these two methods can effectively improve the performance of the SSD model.The accuracy of target detection is thus optimized for the SSD model.1.The 1×1 feature map generated by Conv11＿2 in the SSD prediction layer is up-sampled using double-line interpolation to generate a 3x3 feature map and convolve the feature map.The 3×3 feature map generated after convolution is compared with conv10＿2 Perform feature fusion as the 3×3 feature map in the prediction layer,and the remaining feature maps use the same method described above for upsampling and feature fusion to sequentially generate 6 feature maps in the SSD prediction layer,and then use these 6 feature maps to construct features Pyramid and detect targets.The experimental results show that the accuracy of the proposed method reaches 85.6%,which is 5.4% higher than the accuracy of the original algorithm.2.Insert the CBAM module into the two low-level convolutional layers Conv4＿3 and Conv7 with rich semantics in the SSD model,and then use the improved model to detect the data set.At the same time,the CBAM module is inserted into the SSD model with the feature pyramid,and the CBAM module is inserted after the feature fusion convolution layer Conv4＿3 and Conv7,and the new model is tested.The experimental results show that the accuracy of these two methods are 89.8% and 90.6%after inserting the CBAM module,which is 9.6% and 10.4% higher than the original algorithm,which proves that the performance of the model has been improved,and the effectiveness of the method is proved.