An algorithm combining principal component analysis (PCA)-linear discriminant analysis (LDA) with support vector machine (SVM) is proposed for the real-time and robustness requirements of multi-branch path identification and tracking in the process of automated guided vehicle(AGV) visual guidance. Firstly, the image features are obtained by image preprocessing algorithm and PCA-LDA algorithm. Next, the image is identified by SVM classifier, which is optimized by gray wolf optimization algorithm. In the aspect of path tracking, the lateral deviation and course deviation are calculated by using the least square fitting method. The experimental results show that the rate of path recognition is 99.3% and real-time requirements are achieved by using the algorithm combining PCA-LDA with SVM, and the path tracking error is within 20 mm to meet the general industrial environmental needs.