Human targets in thermal infrared images are easy to observe and have a wide range of applications. However, they are limited by the hardware of thermal infrared devices. The edges of human targets in the images are often blurred and the detection efficiency is poor. Simultaneously, because of the special imaging principle of thermal infrared, human target detection is vulnerable to the interference of heating and occlusion objects and the detection accuracy cannot be guaranteed. In response to the above issues, this study proposes a type of holistically nested edge detection (HED)-thermal infrared saliency human detection network. The network adopted the form of a similar HED network and detected human targets by adding the residuals of different proportions of the hole convolutional codec module. Experiments showed that the network can effectively detect human targets, accurately predict the edge structure, and also have high detection accuracy in an environments with heating objects and obstructions.