基于S^3FD的室内头部检测方法

Indoor Head Detection Method Based on S^3FD

为实现室内小尺度、密集、模糊的头部检测,提出一种基于S^3FD的室内头部检测方法。利用ATSS算法改进先验框匹配策略,增加模型性能,同时加入SAPD算法修改位置损失函数,改善注意力偏差问题。实验结果表明,该方法能够准确定位室内头部目标,有较高的检测精度,可以用于现实应用场景模块构建。

To achieve indoor small-scale,dense and fuzzy head detection,an indoor head detection method based on S^3FD is proposed.The ATSS algorithm is used to improve the prior frame matching strategy and increase the model performance.At the same time,SAPD algorithm is added to modify the position loss function to improve the problem of attention deviation.Experimental results show that this method can accurately locate indoor head targets,has high detection accuracy,and can be used to construct modules in real application scenarios.