Abstract: A lightweight infrared image pedestrian detection algorithm (RT_YOLOv8) based on YOLOv8 was developed to solve the problems of the low fault tolerance rate and large network model of infrared image pedestrian recognition. First, RepGhostNet was used as the backbone network to improve the overall feature extraction ability of the model for pedestrians. The triplet attention mechanism was introduced to improve the ability of the network to extract the target feature information through the three-dimensional interaction of the width (W), height (H), and channel (C). The deformable convolutional network version 2 (DCNv2) was used to reconstruct the C2f module and improve the detection ability of the network. PConv was added to the detection module to replace a portion of the conventional convolution, and reduce redundant computing and memory access, and the spatial features were effectively extracted. The results of an experiment showed that, compared with YOLOv8, the RT_YOLOv8 pedestrian detection algorithm increased mAP50 by 1.51% and decreased GFLOPS by 50.62% on the infrared dataset established in the experiment, which met the deployment requirements of low computing power and high precision.