A traffic sign text detection algorithm (efficient traffic sign text detection, ETST) based on the efficient and accurate text detection (EAST) algorithm was proposed in order to address the issue of poor accuracy and stability in traffic sign text detection in highway scenarios. The detection performance of highway traffic sign text was improved by introducing three types of modules. Potential marker regions in an image can be quickly screened by utilizing a traditional marker pre-detection module based on color threshold segmentation and shape analysis. The recognition ability of the proposed algorithm for text at different scales was improved by integrating a multi-scale attention enhancement module. An improved parallel patch-aware attention module (NPPA) was proposed to enhance the ability of the proposed algorithm to perceive and represent local structural features. ETST effectively combined low-level texture cues with high-level semantic information by fusing multi-level feature maps in order to enhance discrimination of challenging text instances. Experiments on datasets such as ICDAR2015, CCTSDB and Total-Text demonstrated that ETST outperformed existing algorithms in key metrics such as accuracy, recall and F₁ score, while achieving a processing speed of over 10 frames per second. Then the requirement of near real-time application was satisfied, and decent detection accuracy and computational efficiency were confirmed.