基于卷积神经网络的实时行人检测方法

摘要

In recent years,the convolution neural networks in the field of pedestrian detection have achieved similar and even better results,compared to other methods.However,the slow detection speed can't meet the realistic demand.To solve this problem,a real-time pedestrian detection method is put forward.The scattered detection processes are integrated into a single depth network model.Images which can be calculated through the model can directly output detection results.The extended ETH dataset is used for training and testing the model.The experimental results show that the method is very fast and can achieve the goal of real-time detection with the guaranteed accuracy.%近年来,卷积神经网络在行人检测领域取得了同其他方法相似甚至更好的检测成绩,然而缓慢的检测速度远不能满足现实需求.针对这一问题,本文提出一种实时的行人检测方法,将分散的检测过程整合成单一的深度网络模型,被检测图片通过模型的计算可以直接输出检测结果.使用扩充的ETH数据集进行训练测试,实验结果表明,在保证准确率的情况下,该方法检测速度极快,可以满足实时检测的目的.