Abstract: A multi-classification algorithm for indoor positioning based on SVM was proposed to tackle the problem of low precision and fluttering results faced in many real-time location systems. Traditional matching algorithms based on sampling points arc always deficient in dealing with nonlinear problem and jumping results in a short time. In handing this limitation,object location process was considered as a multi-classification problem by introducing grid concept K candidate grids were obtained using SVM first These candidates were then refined by previous location results, and ultimate accuracy result was achieved through a Kalman filter. Temporal information was utilized in the matching process to make the object movement more stable and smooth. Experiments show the superiority of our method over naive SVM method.

Key words: Support vector machine(SVM),Urid,Real-time indoor location,Received signal strength indication(RSSI)，Kalman filter