Font Size: a A A

Research On Road Scene Object Detection Algorithm Using Deep Learning

Posted on:2021-03-01Degree:MasterType:Thesis
Country:ChinaCandidate:Y H CuiFull Text:PDF
GTID:2428330629487221Subject:Control Engineering
Abstract/Summary:PDF Full Text Request
Road object detection technology is widely used in unmanned driving,smart transportation and other fields,and has received widespread attention from academia and industry.However,due to the characteristics of various types of road traffic,overlapping of multiple targets,and variable climate,how to efficiently classify and locate road targets is a challenge in smart transportation.Traditional object detection algorithms are mostly based on manually designed features.Although they can detect objects,they have many defects.In recent years,deep learning-based object detection methods use convolutional neural networks to automatically extract the features of the object,and then detect the objects in the image,which greatly improves the detection accuracy and speed.In this paper,based on the characteristics of road targets,on the basis of systematically summarizing the existing target detection algorithms,with the goal of improving detection accuracy and speed of the algorithm,the research on road target detection algorithms in complex scenarios is carried out.In order to reduce the computational complexity of the model and speed up the detection speed of the model,this paper improves the lightweight neural network MobileNet,and then combines with the YOLO V3 target detection algorithm to propose a lightweight road target detection model——MobileNet-YOLO.The network embeds SEnet as a module in MobileNet to learn the interdependence of each feature channel.On this basis,the spatial pyramid pooling algorithm is introduced to solve the problem of different input image sizes of the network.Finally,the improved MobileNet is used as the feature extraction network to replace the Darknet53 in YOLO V3.The experimental results of the VOC dataset show that the proposed MobileNet-YOLO detection model greatly improves the speed of target detection without losing too much accuracy.Although the detection speed of MobileNet-YOLO is faster than the original YOLO V3 algorithm,in order to further improve the detection accuracy of the road target detection algorithm,a road target detection model based on the improved YOLO V3 algorithm is proposed.This algorithm improves the YOLO V3 algorithm.It selects and combines multiple scale output feature maps to make predictions,which improves the model's ability to detect small targets and occluded objects.The K-means clustering algorithm selects the best pre-selection box.The size and quantity,and at the same time randomly input pictures of different resolutions into the network,improve the stability of the network to detect targets of different sizes,and improve the detection effect of the network model.Finally,through experiments,it is verified that the network model designed in this paper is 7%(mAP)higher in accuracy than the original YOLO V3,and has good engineering application value.
Keywords/Search Tags:object detection, convolutional neural network, MobileNet, SEMet
PDF Full Text Request
Related items