Active millimeter wave imaging ( AMWI) is an efficient way to detect dangerous objects concealed under clothes. However, because the images acquired by AMWI are often obscure and some of concealed objects are small in size, the automatic detection and localization of the objects remain as a challenging problem. Yao[1]first employed convolutional neural networks ( CNNs) and used a dense sliding windowmethod to detect concealed objects. In this paper, the author presents two improvements over Yao 's work: 1) Using contextual information to suppress interference and improve detection probability;2) Using a two-step search method instead of exhaustive search to reduce the computational complexity. To reduce the computational complexity, the author first uses a CNN in vertical direction to filter the interference and obtain the vertical position of the concealed object, then uses another CNN to determine the horizontal position of the concealed object. To make use of big windowcontaining contextual information, the author uses IoG ( intersection-over-ground-truth) instead of IoU ( Intersection-over-Union) to define positive and negative samples in training and testing process. Experimental results showthat the proposed method will make the length of computational time reduced to about 30％ of that of the exhaustive search while achieving better detection performance.%主动式毫米波成像 (AMWI) 技术是检测隐藏在衣服下的危险物体的有效方法.但AMWI获取的图像通常很模糊, 而且一些隐匿物体的尺寸较小, 因此隐匿物品的自动检测和定位仍然是一个具有挑战性的问题.姚家雄等[1]首先使用卷积神经网络 (CNNs) 结合穷举滑动窗口方法来检测隐藏物体.做了两点改进: (1) 使用上下文 (背景) 信息抑制干扰, (2) 使用两步搜索方法代替穷举搜索来降低计算复杂度.首先在垂直方向上使用一个CNN来过滤干扰, 得到隐藏物体的垂直位置, 然后用另一个CNN来确定水平位置.为了充分利用上下文信息, 使用IoG (交集和真值的比) 代替IoU (交并比) 来定义训练和测试过程中的正负样本.实验结果表明, 该方法将计算时间减小到约30％, 同时实现更好的检测性能.
展开▼
