基于改进卷积神经网络的动态障碍物检测方法
|
摘要
针对传统卷积神经网络在感兴趣目标较小的情况下对动态障碍物检测会出现结果不准确的问题,提出一种基于改进卷积神经网络的运动障碍物检测方法。该方法在层层抽取细节特征的基础上融入全局特征,利用全局特征修正细节特征的提取,利用Softmax层进行分类来获取图像的整体信息。实验结果表明,与传统卷积神经网络相比,改进卷积神经网络具有较低的时间复杂度,以及较高的识别率。
Traditional convolution neural network can produce inaccurate results in dynamic obstacle detection when the target of interest is small.To solve this problem,we proposed moving obstacle detection method based on improved convolution neural network.The method integrated global features into detail feature extraction layer by layer.We used the global features to modify the extraction of detail features,and used the Softmax layer to classify to obtain the overall information of the image.The experimental results show that the improved convolution neural network has lower time complexity and higher recognition rate than the traditional convolution neural network.
Traditional convolution neural network can produce inaccurate results in dynamic obstacle detection when the target of interest is small.To solve this problem,we proposed moving obstacle detection method based on improved convolution neural network.The method integrated global features into detail feature extraction layer by layer.We used the global features to modify the extraction of detail features,and used the Softmax layer to classify to obtain the overall information of the image.The experimental results show that the improved convolution neural network has lower time complexity and higher recognition rate than the traditional convolution neural network.
引文
[1]Zhou H P,Liu Q B,Chen D P,et al.Doppler Shift and Height Detection of Obstacle Based on FMCW Radar Sensor[C]//2015 International Conference on Cyber-Enabled Distributed Computing and Knowledge Discovery(Cyber C).IEEE,2015:399-402.
[2]邹斌,刘康,王科未.基于三维激光雷达的动态障碍物检测和追踪方法[J].汽车技术,2017(8):19-25.
[3]金立生,王岩,刘景华,等.基于Adaboost算法的日间前方车辆检测[J].吉林大学学报:工学版,2014,44(6):1604-1608.
[4]Michael D,Nagasaka N,Kamata N,et al.Classifying and passing 3D obstacles for autonomous driving[C]//IEEE International Conference on Intelligent Transportation Systems.IEEE,2015:1240-1247.
[5]王铮,赵晓,佘宏杰,等.基于双目视觉的AGV障碍物检测与避障[J].计算机集成制造系统,2018,24(2):400-409.
[6]Tian Y,Luo P,Wang X,et al.Pedestrian detection aided by deep learning semantic tasks[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).IEEE,2015:5079-5087.
[7]Ahmed E,Jones M,Marks T K.An improved deep learning architecture for person re-identification[C]//Computer Vision and Pattern Recognition.IEEE,2015:3908-3916.
[8]张永良,张智勤,吴鸿韬,等.基于改进卷积神经网络的周界入侵检测方法[J].计算机科学,2017,44(3):182-186.
[9]Brezak M,Petrovic I.Real-time approximation of clothoids with bounded error for path planning applications[J].IEEETransactions on Robotics,2014,30(2):507-515.
[10]刘万军,梁雪剑,曲海成.不同池化模型的卷积神经网络学习性能研究[J].中国图象图形学报,2016,21(9):1178-1190.
[11]Le Cun Y,Bottou L,Bengio Y,et al.Gradient-based learning applied to document recognition[J].Proceedings of the IEEE,1998,86(11):2278-2324.
[12]朱爱斌,刘洋洋,何大勇,等.解决路径规划局部极小问题的势场栅格法[J].机械设计与研究,2017(5):46-50.
[13]钱基德,陈斌,钱基业,等.基于感兴趣区域模型的车道线快速检测算法[J].电子科技大学学报,2018,47(3):356-361.
[14]Demetz O,Hafner D,Weickert J.Morphologically invariant matching of structures with the complete rank transform[M].Kluwer Academic Publishers,2015.
[15]Cordts M,Omran M,Ramos S,et al.The cityscapes dataset for semantic urban scene understanding[EB].eprint ar Xiv:1604.01685,2016.
[2]邹斌,刘康,王科未.基于三维激光雷达的动态障碍物检测和追踪方法[J].汽车技术,2017(8):19-25.
[3]金立生,王岩,刘景华,等.基于Adaboost算法的日间前方车辆检测[J].吉林大学学报:工学版,2014,44(6):1604-1608.
[4]Michael D,Nagasaka N,Kamata N,et al.Classifying and passing 3D obstacles for autonomous driving[C]//IEEE International Conference on Intelligent Transportation Systems.IEEE,2015:1240-1247.
[5]王铮,赵晓,佘宏杰,等.基于双目视觉的AGV障碍物检测与避障[J].计算机集成制造系统,2018,24(2):400-409.
[6]Tian Y,Luo P,Wang X,et al.Pedestrian detection aided by deep learning semantic tasks[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).IEEE,2015:5079-5087.
[7]Ahmed E,Jones M,Marks T K.An improved deep learning architecture for person re-identification[C]//Computer Vision and Pattern Recognition.IEEE,2015:3908-3916.
[8]张永良,张智勤,吴鸿韬,等.基于改进卷积神经网络的周界入侵检测方法[J].计算机科学,2017,44(3):182-186.
[9]Brezak M,Petrovic I.Real-time approximation of clothoids with bounded error for path planning applications[J].IEEETransactions on Robotics,2014,30(2):507-515.
[10]刘万军,梁雪剑,曲海成.不同池化模型的卷积神经网络学习性能研究[J].中国图象图形学报,2016,21(9):1178-1190.
[11]Le Cun Y,Bottou L,Bengio Y,et al.Gradient-based learning applied to document recognition[J].Proceedings of the IEEE,1998,86(11):2278-2324.
[12]朱爱斌,刘洋洋,何大勇,等.解决路径规划局部极小问题的势场栅格法[J].机械设计与研究,2017(5):46-50.
[13]钱基德,陈斌,钱基业,等.基于感兴趣区域模型的车道线快速检测算法[J].电子科技大学学报,2018,47(3):356-361.
[14]Demetz O,Hafner D,Weickert J.Morphologically invariant matching of structures with the complete rank transform[M].Kluwer Academic Publishers,2015.
[15]Cordts M,Omran M,Ramos S,et al.The cityscapes dataset for semantic urban scene understanding[EB].eprint ar Xiv:1604.01685,2016.