Aiming at the problem of corner information extraction in automatic recognition and location of key points to be measured in machine vision measurement applications, a real-time video image corner detection system is built based on the heterogeneous architecture of ARM+FPGA in ZYNQ series expandable platform, using hardware and software co-design method. Using Vivado HLS tools, corner detection algorithm is encapsulated into an IP core that can be deployed to the PL end, which greatly shortens the development cycle. Reasonable task allocation is made for each module of the system, which makes the system have the flexibility of ARM and the parallel processing ability of FPGA, and shows the advantages of parallel heterogeneous architecture. The IP core of image algorithm in the system can be replaced and updated flexibly, which provides an important reference for miniaturized application based on machine vision detection.