Abstract: According to the large sampling workload for traditional fingerprint localization algorithms in the offline phase, a partition fitting approximation method (P-FP) is proposed to construct the offline fingerprint database. To solve the problem of the low location accuracy caused by the fluctuation of the WiFi signal in the online phase, the PS-FP algorithm, a sampling importance resampling (SIR) particle filter with threshold based P-FP, is further proposed. The entire indoor parking lot is partitioned, and the environmental coefficients of each partition can be gained by using the polynomial fitting model. And the error characteristic matrix is established with the difference between the fitting values and the actual measured values. Then, the virtual reference points (RPs) is redeployed, and C-means clustering is utilized for the offline fingerprint database. The optimal threshold is obtained by comparing the mean location error (MLE), then the PS-FP is used to optimize the online location coordinates. Experimental results demonstrate that PS-FP can achieve high location accuracy when the RP is fewer, and the mean location error is only about 0.7 m. Cumulative distribution function (CDF) shows that 98% of the location errors are within 2 m obtained by PS-FP.