Lightning disasters pose a serious threat to railway operations due to their strong spatiotemporal randomness and high destructiveness. Traditional lightning location methods rely on manual inspections, which suffer from difficulties in detection and delayed responses. To address the aforementioned challenges, a computer vision-based positioning method was proposed that integrated adaptive background subtraction with enhanced edge detection, specifically designed for complex railway scenes and overexposure issues in lightning imaging. This approach employed a dynamic Gaussian mixture model for background modeling, combined an improved Canny operator to enhance lightning channel feature extraction, and incorporated morphological filtering to eliminate noise interference, thereby establishing an efficient framework for lightning target detection. To verify the effectiveness of the method, 37 sets of natural lightning cases with different scenarios were selected for experimentation. The results show that the recall and accuracy of lightning channel recognition are 0.86 and 0.89, respectively, and the accuracy of lightning landing point recognition reaches 97.3%. The proposed method significantly improves the robustness and real-time performance of lightning localization in complex environments, effectively alleviates the problem of feature loss caused by overexposure, and has high accuracy in lightning landing point localization. It can provide reliable technical support for intelligent monitoring and rapid emergency response of railway lightning disasters.