卷积神经网络的多尺度行人检测

Multi-scale pedestrian detection based on convolutional neural networks

行人检测在智能监控、自动驾驶、辅助驾驶、智能机器人等研究领域有着广泛的应用.传统的行人检测方法大多使用滑动窗口遍历图片的方式,导致计算量大,检测速度受到限制.目前基于深度学习的行人检测方法进入了一个快速的发展阶段,但是还存在例如小尺寸行人漏检等很多问题.现提出基于卷积神经网络的多尺度行人检测方法,分析了增加检测层、并联卷积层与改变卷积核尺寸对行人检测性能的影响.在KITTI数据集上的实验结果表明,该方法可以实现较好的行人检测效果.

Pedestrian detection has been widely applied in intelligent surveillance,automatic driving,driver assistance systems,intelligent robots and so on.The traditional pedestrian detection method,by using the moving window traveling over images,leads to heavy computational cost and low speed of pedestrian detection.At present,the pedestrian detection method based on deep learning has entered into a stage of rapid development.But there are still many problems,such as the missing detection of pedestrians in small size.In this paper,we proposed a multi-scale pedestrian detection method based on convolutional neural networks.We analyzed some factors such as the increase of detection layers,parallel convolution layers,changes of the size of convolution kernels,and their impact on the performance of pedestrian detection.The experimental results show that the proposed method can perform well on KITTI data sets.