Travel has always been a problem for blind people due to their lack of visual perception of the surrounding environment. This paper presents an improved obstacle detection algorithm CE-YOLOX based on YOLOX for guide system. Firstly, in order to reduce semantic information loss caused by feature fusion network when feature channel is reduced, sub-pixel hopping fusion module SSF and sub-pixel context enhancement module SCE are used to make full use of channel information and semantic information of different scales, and channel attention guide module CAG is used to reduce aliasing effect caused by multi-scale feature fusion. Secondly, in order to make the model more focused on effective features, the global attention mechanism GAM is introduced to improve the performance of the model by reducing the information dispersion and amplifying the global interactive representation. Then, the position regression function IOU-LOSS of the original model is replaced by SIOU-LOSS, which speeds up the regression speed and precision of the frame. Finally, the detection platform of the guide system is built and the proposed algorithm is transplanted to the edge computing device NVIDIA Xavier NX. The experimental results show that the obstacle algorithm of the improved guide system has the same mAP on the server and NVIDIA Xavier NX platform, which is improved to 90.53%, 2.45% higher than the original YOLOX model algorithm. The detection speed reaches 75.93 FPS on the server. The model in this paper not only gives consideration to the detection speed but also improves the accuracy, which is significantly better than the comparison algorithm. It meets the requirements of edge computing equipment and has practical application value.