MOVING CROWD COUNTING BY INTERGRATING SINGLE AND MULTIPLE COLUMN NEURAL NETWORK
Existing crowd counting methods are limited to counting the integrity of the crowd,the accuracy rate is downgraded when exclusively counting the moving people in the crowd.An attention based multi-stage deep learning framework is proposed to solve this problem.Attention module was adopted to adaptively selects both single-column and multi-column counting networks,combine the deep features of single column network and the multiple scale receptive fields of multiple column network,which effectively extracted features of the moving people.The results show that the proposed method has lower mean square error(MSE)and mean absolute error(MAE)than existing crowd counting methods.The counting accuracy of people on moving is well improved.
Crowd countingDeep learningSingle and multiple column networkAttention mechanism
国家科技期刊平台
NSTL
万方数据
