A high accurate optical head pose tracker system based on multiple-cameras was investigated to track the high accurate head pose of a virtual reality system. A calibration square was designed to eliminate the system installation error to allow the optical axis of a camera to be perpendicular every two cameras.In positioning, the depth value of the object was given arbitrarily by taking a camera for a criterion, then, other coordinate values could be generated by a camera model. The obtained results were taken as the initial depth and used in the loop iteration to compute the object's position. At last, the object pose could be calculated based on the position of three markers. Comparative experiments show that the algorithm has high accuracy and rapid convergence. The static error is 0.51 mm, and the dynamic tracking error is 0.88 mm,which is significantly higher than that of the electromagnetic tracker system. Meanwhile, the positioning system has low-cost and can not be disturbed by metal or electromagnetic environments. It can meet the high-precision position tracking requirements of virtual reality systems.