Research On Target Detection Of Unmanned Vehicle Based On Deep Learning

In the driverless technology,through the target detection technology,accurately and quickly identify the target,so that vehicles through mutual coordination and cooperation,can effectively avoid traffic congestion and traffic safety problems.In the automatic driving environment,due to the complexity of the target and the variety of occluded or small targets,the automatic driving target detection needs high accuracy and real-time performance,in order to ensure the safety of road drivers and pedestrians and the efficiency of travel.In order to make the model fast and accurate target detection,the following research is made on the basis of SSD algorithm1)Using MobileNet as SSD skeleton network based on depth decomposable convolution,the SSD＿MobileNet model.In convolution network,with the deepening and complexity of convolution network,the target semantic information will be more fully expressed,but the detection speed will be greatly reduced on the embedded platform of unmanned driving.The experimental results show that,due to the depth decomposable convolution of mobilenet model,the calculation amount can be effectively reduced,and the size of the model can be reduced.Compared with the original SSD model,the detection accuracy is guaranteed and the detection speed is accelerated.2)In order to improve the detection accuracy,especially for small targets or occluded targets,On the basis of SSD＿MobileNet model,specific feature layers are selected.Each feature layer recursively performs bidirectional feature fusion after core module and purification module to form residual bidirectional fusion feature pyramid structure,FSSD＿MobileNet network circularly flows rich semantic information in deep layer and rich location information in shallow layer,which makes feature extraction more sufficient.The experimental results show that FSSD＿Mobilenet network has higher detection accuracy for small targets and occluded targets,and solves the problem of missing detection of small targets and occluded targets to a certain extent.Figure 47;Table 15;Reference 55...