A certain sensor error in multi-sensor target tracking may influence the target track precision. In order to solve this problem, adopt least squares curve fitting and kalman one-step prediction, sensor which temporarily without auto-identity function will not use the polluted data to make the output data more precocious and system stronger. Then, adopt interacting multiple model (IMM) to carry out simulation test. The test result shows that the algorithm is simple, efficient and with good prospect of engineering application.