基于ResNet-LSTM的行人过街行为识别方法

Pedestrian Crossing Behavior Recognition Method Based on ResNet-LSTM

为提高行人行为识别精度,通过基于光流处理的Resnet-LSTM网络模型对行人过街行为进行识别。在采用光流法对连续视频帧进行处理的基础上,基于ResNet神经网络提取有序光流数据信息的空间维度特征,并利用LSTM网络进行时序性分析,完成行人过街行为的分类识别。论文利用Weizmann数据集对该模型做有效性检验,结果表明,算法的行为识别率可达99.46%。

In order to improve the accuracy of pedestrian behavior recognition,the pedestrian crossing behavior is identified by Resnet-LSTM network model based on optical flow processing.On the basis of continuous video frame processing by optical flow method,spatial dimensional characteristics of ordered optical flow data are extracted based on ResNet neural network,time series analysis is completed by using LSTM network,the classification and recognition of pedestrian crossing behavior are accomplished.This paper uses Weizmann data set to test the validity of the model.The results show that the recognition rate of the proposed algorithm can reach 99.46%.