改进的SSD算法在智慧交通中的应用

针对智慧交通管理系统中交通车辆监控、车流量统计、违法车辆追踪等问题，为了提高目标车辆检测的准确率和效率，提出了一种改进的SSD(Single Shot MultiBox Detector)目标检测算法。该算法将相邻的卷积层进行特征信息融合，提高准确率；通过减少部分卷积层的深度，提高计算效率；为了提高泛化能力，在减少1×1特征图的情况下对有效特征层增加归一化批处理后进行分类预测。改进的SSD目标检测算法能很好地检测出复杂场景下的多类交通工具，并且在训练过程中训练步数减少，正确率更高。实验结果表明，所提的改进的SSD算法在不改变VGG16图片大小的情况下，平均检测精度比原有的SSD算法提高了15.59%,同批次测试数据用时更短并且收敛速度更快，达到了预期效果。

Application of improved SSD algorithm in intelligent transportation

For the problems of traffic vehicle monitoring,traffic flow statistics and illegal vehicle tracking in intelligent traffic management system,an improved Single Shot MultiBox Detector(SSD)target detection algorithm is proposed to improve the accuracy and efficiency of target vehicle detection.The algorithm fuses the feature information of adjacent convolution layers to improve the accuracy.By reducing the depth of partial convolution layer,the computational efficiency is improved.In order to improve the generalization ability,1×1 feature maps are reduced and the effective feature layer is classified and predicted after adding normalization batch.The improved SSD target detection algorithm can detect many kinds of vehicles in complex scenes,and reduce the number of training steps in the training process with higher accuracy.The experimental results show that the improved SSD algorithm improves the mean average precision value by 15.59%compared with the original SSD algorithm without changing the size of VGG16 image,shortens the time of same batch of test data,improves the convergence speed,and achieves the desired effect.