Research On Road Target Detection Technology Based On Embedded

As a modernized transport artery,the highway has important research value which plays an important role in the development of the economy.A road target detection system was put forward in the paper,focusing on the increasingly frequent traffic accidents on structured roads.The detection system is used to improve the intelligent level,to realize the intelligent detection of road targets,to reduce traffic accidents and to improve the traffic capacity of the highway.In road targets,the amount of information represented by lane lines and license plates is very rich,which is of great significance.Therefore,based on embedded road target detection system,it realizes lane line detection and vehicle license plate recognition.Lane line detection includes three parts: image preprocessing,lane line feature extraction and lane line tracking.The lane line feature detection proposed in the paper includes edge feature extraction,color feature extraction,and lane line feature extraction based on the gauss filter operator based on lane width and direction.After the lane line feature extraction,the lane line model is proposed to improve the accuracy of lane detection and based on spatiotemporal consistency,a lane line tracking algorithm is proposed which increases the lane detection speed and improves the real time performance of the system.License plate recognition includes two parts: license plate detection and license plate character recognition.In the paper,the Haar classifier is used to detect the area of the license plate,and the area of the license plate is accurately located.The orthographic view of the exact area of the license plate is obtained through the perspective transformation.Then the segmentation of the character of the license plate is realized by using the histogram matching method of the dynamic programming and the character segmentation algorithm of the CNN(Convolutional Neural Network),and the segmentation character is identified by CNN.So,the result of the license plate recognition is obtained.The paper study two main targets in the road image which are lane line and license plate.At the same time,the road targets are detected based on the FCN8(Full Convolution Network8)model and the PSPNet(Pyramid Scene Parsing Network)model,then realized the semantic deconstruction of road scene.The FCN8 model can identify 16 kinds of road targets,and the PSPNet model can identify 33 road targets.In the study,the road target detection system is transplanted to the Cortex-A53 development board,realizing the road target detection of the embedded platform.The road image size is 1944×2592,the detection rate of the lane line detection is 26 frames per second,its accuracy reaches 92.8%,the license plate recognition accuracy reaches 97.2%,and the detection rate of the road target detection system is 24 frames per second.The road target detection system can intelligently manage vehicles and detect illegal vehicles,so it has wide application value.