Indoor localization has become a foundational capability for a wide range of emerging applications, from asset tracking and logistics in industrial settings to context-aware services in retail, healthcare, and smart buildings. Currently, there are problems such as inaccurate positioning accuracy and high measurement costs in indoor positioning. Therefore, we have proposed a positioning method based on the existing WIFI framework that supports the FTM protocol. However, in practical indoor environments FTM measurements are often corrupted by multipath and non-line-of-sight (NLOS) propagation. These effects introduce systematic biases and heterogeneous noise across access points (APs), which substantially degrade positioning accuracy. Therefore, we propose an error modeling plus weighted least-squares (WLS) localization method driven by RSSI and FTM fusion to improve the positioning accuracy: for each AP, RSSI + log(FTM) are used to classify LOS/NLOS based on GMM and SVM, and class-conditional error mean and variance are estimated. The per-class mean is subtracted to bias-correct FTM ranges, and the inverse variance is used as a dynamic weight; the top three APs by RSSI are selected and a height-constrained WLS is solved. In three different environments including lobby, classroom and dormitory, the proposed method achieves mean absolute errors (MAE) of 1.09 m, 0.68 m and 1.49 m, respectively, substantially outperforming traditional baselines.