Camouflage people detection via strong semantic dilation network

Camouflage greatly reduces the probability of the target being discovered, detecting the camouflage object from background is very hard for former object detection methods because they are designed to detect the object in an ideal environment. It brings difficulties to subsequent applications such as battlefield information acquisition, precision guided weapons and object tracking. To deal with the problem, in this paper, a new complete camouflage people detection dataset is firstly constructed, then we propose strong semantic dilation network (SSDN), which is specially designed to detect camouflage people in an end-to-end architecture. SSDN makes full use of semantic information in CNN and dilated convolutions are also added to enlarge the receptive field to find camouflage people. Experiments demonstrate that SSDN perform well in camouflage people detection dataset compared with other object detection methods.