<p>Rapid localization of shield tunnel segments based on 3D laser scanning is a prerequisite for tunnel dislocation detection. To address this issue, this paper proposed a training-free and sample-independent joint identification algorithm based on improved genetic algorithm optimization. The algorithm can achieve accurate segment localization and also serve as an auxiliary tool for sample annotation. By introducing elite retention, population diversity enhancement, and early stopping strategies, the global search capability and convergence efficiency were significantly improved. In joint identification, the robustness and generalization ability of the algorithm in complex scenarios were improved by optimizing the denoising threshold introduced. Finally, a dislocation detection algorithm based on the 3D point cloud unfolding was proposed. The results show that the improved genetic algorithm reduces computing time by over 40% and average error by more than 5% compared with the average performance of other optimization algorithms. The joint identification algorithm achieves false negative and false positive rates for both circumferential and longitudinal joints within 10%, with average deviations of 4.31&#xa0;mm and 5.38&#xa0;mm, improving by 35% over the edge detection + Hough line method, and exhibiting stronger generalization than the deep learning algorithm. Using the joint identification algorithm as an auxiliary tool for sample annotation, compared with the traditional manual annotation method, a speed improvement of 1.38 times has been achieved. Longitudinal joint and circumferential joint dislocation average deviations are 0.72&#xa0;mm and 0.88&#xa0;mm, over 50% more accurate than traditional 2D elliptic-fitting-based dislocation detection.</p>