Traditionally, stereo vision was mainly based on FPGA+CPU method, which was limited by lower computing capability of CPU. An innovative all-hardware acceleration solution of stereo vision is proposed for positioning application, in which image acquisition, caching, preprocessing and 3D positioning modules are all implemented in FPGA. A novel median filter for image preprocessing is proposed to match higher clock frequency. A complex 3D positioning algorithm is given to simplify the process without reasonable reduction in operation accuracy, which makes the system more conducive to implement in FPGA. The experiments show that the all-hardware acceleration solution has outstanding advantages in computing speed, power consumption and response time compared with the traditional FPGA+CPU one.