To solve the stereo matching and measurement accuracy problems which exist in stereo vision systems in the field of robot environment perception, this paper designs a stereo vision system based on scale-invariant feature transform(SIFT) features. The matching problem is solved by using the rotation, scale, illumination invariance of the SIFT features, and the SIFT algorithm is implemented in the embedded system based on FPGA and DSP, thus the real-time problem of the system is effectively solved. Then in order to offset the problem that the measurement error increases with the distance, an error compensate method based on quadratic polynamial is proposed, which can compensate the measurement results of the stereovision system and improvus the measurement accracy of the system. The mesurement experiment, robot environment perception experiment and contrast experiment between existing stereovision products are uarried out, and the uasults show that, the stereovision system can commendably solve the matching and accuracy problems, and also can be better applied to the environment perception task of mobile robots.