期刊文献+

基于Triangle Net的密集人群计数 被引量:1

下载PDF
导出
摘要 文章提出了一种编码解码器网络Triangle Net。编码器生成多分辨率特征,解码器中密度估计模块与编码器的输出并行连接,使得每一个特征图都能充分融合其他连接中的有效信息以生成最佳密度图;加固模块对编码器输出进行了连续上采样以加速网络收敛,最后融合上述两模块的人群密度图进行高精度人群计数。实验结果表明本方法有效降低了平均绝对误差(MAE)和均方误差(MSE),证明该模型在密集人群计数上有较好准确度和鲁棒性。 In this paper,a kind of codec network Triangle Net is proposed.The encoder generates multi-resolution features,and the density estimation module in the decoder is connected with the output of the encoder in parallel,so that each feature map can fully integrate the effective information in other connections to generate the best density map;the reinforcement module carries out continuous up-sampling of the encoder output to accelerate network convergence,and finally fuses the crowd density map of the above two modules for high-precision crowd counting.The experimental results show that the method effectively reduces the mean absolute error(MAE)and the mean square error(MSE),which proves that the model has good accuracy and robustness in dense crowd counting.
出处 《科技创新与应用》 2021年第9期38-40,44,共4页 Technology Innovation and Application
关键词 人群计数 卷积神经网络 多尺度融合 并行连接 加固模块 crowd counting convolution neural network multi-scale fusion parallel connection reinforcement module
  • 引文网络
  • 相关文献

同被引文献4

引证文献1

相关主题

;
使用帮助 返回顶部