Augmented reality (AR)-assisted maintenance has emerged as a transformative approach to enhance the efficiency of traditional manual fault diagnosis in automotive rear axle assembly lines. However, the weak-textured surfaces and structural complexity of industrial equipment pose significant challenges to stable 3D tracking registration, hindering the practical deployment of AR systems. This study proposes a hybrid framework integrating an improved LineMod algorithm with natural feature-based techniques for virtual-real fusion in weak-textured environments. The framework comprises two modules: an offline phase for multi-view template generation using a uniformly sampled icosahedron-based spatial grid, and an online phase combining gradient-based LineMod template matching with ORB-KLT optimization. Gradient descriptors and diffusion strategies are employed to enhance template matching robustness against illumination variations and partial occlusions. Subsequently, ORB feature extraction and KLT optical flow tracking are synergistically applied to refine camera pose estimation and improve computational efficiency. Experimental evaluations demonstrate that the proposed method achieves high precision, real-time performance, and robust initialization in weak-textured industrial scenarios, effectively addressing the challenges of 3D tracking registration for AR-assisted maintenance applications. Moreover, comparative tests revealed superior robustness against occlusions compared to original LineMod algorithm.