The wearable localization system for wireless capsule endoscopy (WCE) is a potential technology to realize rapid diagnosis and treatment of the gastrointestinal (GI). However, the electromagnetic localization accuracy of WCE still needs to be improved. In this paper, based on RSSI electromagnetic fading model, the accurate fitting parameter values are obtained by Kalman filter and the least square method. A novel weighted centroid localization (WCL) algorithm based on exponential weights is proposed, which can achieve high-accuracy localization by using only sparse reception matrix. The simulation results show that when the standard deviation of the localization data is 7.85, the localization root mean square error (RMSE) is 25.4 mm; when the standard deviation of the localization data is 5.475, the localization RMSE is 2.5 mm. These two localization RMSEs are 38% and 79% less than those of the conventional centroid localization algorithm, respectively. An experimental platform of wearable wireless communication and localization system using 24 array receiving antennas is developed in human phantom environment. The experimental results show that the wearable WCE electromagnetic localization system based on the proposed algorithm achieves a localization RMSE of 36.3 mm, which is 17% lower than that of the conventional centroid localization algorithm and meets the needs of clinical diagnosis.