基于改进卷积神经网络的周界入侵检测方法

监控系统已经成为周界入侵防范的重要手段之一,但是目前局限于被动式监视。对此,提出一种通过对监控系统传回的视频图像进行人体目标识别进而实现主动周界入侵检测的方法。针对目前人体目标检测算法场景适用性较差的问题,提出一种基于改进卷积神经网络的行人检测算法,该方法在深层特征的基础上融入浅层特征,利用浅层特征校正深层特征在识别目标过小时的局限性,最后利用Softmax进行分类。实验结果证实,改进后的卷积神经网络对行人的姿态和适用场景都具有较高的鲁棒性,并且在INRIA库上的识别率为98.82%, 在NICTA库上的识别率为99.82%,在CVC库上的识别率为94.50%,在Daimler库上的识别率为99.92%。

Perimeter Intrusion Detection Based on Improved Convolution Neural Networks

Monitoring system has become one of the most important means for perimeter intrusion detection.But most of the existing monitoring systems are passive surveillance.In this paper,a method for active perimeter intrusion detection was proposed by identifying human targets in video images captured by monitoring systems.In order to enhance the robustness of different environment,this paper identified an improved convolution neural networks to realize an effective detection of human bodies with multiple postures captured by fixed cameras.Depth and shallow information are used to describe the pedestrian,so that it can improve the precision and robustness.Then,Softmax is used for classification.The experiment results confirm that the proposed algorithm has higher recognition rate for detecting human targets,which achieves recognition accuracy of 98.82% on INRIA database,99.82% on NICTA database,94.5% on CVC database and 99.92% on Daimler database,respectively.