To address the challenge posed by significant environmental noise that disrupts the parameters of binocular vision systems in pantograph anomaly detection-thereby compromising the accuracy of disparity maps and detection precision-an improved stereo matching method was developed and integrated with anomaly feature detection for this application. This study employs an expanded neighborhood cost calculation and weighted cost aggregation to mitigate the inaccuracies associated with binocular epipolar geometry induced by environmental noise or structural alterations. The local binary pattern method was utilized to compute the cost distribution function， rendering the algorithm applicable for fixed scene detection. Additionally， the disparity search space was constrained to enhance efficiency for real-time trackside detection. Subsequently， abnormal feature detection was conducted on the reconstructed 3D point cloud to derive measurement outcomes. The efficacy of this approach was validated within a pantograph anomaly detection system at railway stations. Experimental results indicate that the proposed algorithm enhances time efficiency by over 30% and measurement accuracy by more than 60% when compared to conventional algorithms， thus effectively addressing the issues of low efficiency and insufficient accuracy in rail transit on-site detection equipment.