The vision measurement system based on a embedded platform was improved. A research algorithm based on dual-threshold in Digital Signal Processing(DSP)was transplant in a Field Programming Gata Arrage(FPGA)to resolve the conflict between target extraction accuracy and data update rate of traditional global search and adjacent domain search. A pixel-level algorithm for cooperative targets was implemented in FPGA on embedded platform. Some functions based on dual-threshold searching in frame data transmission, such as pixel detection, pixel region identification, centroid calculation by a 10-level pipeline, and parallel computing in a multiple marker region were realized. Benefited from this, the contradiction among the neighborhood-size setting, system update rate and the target speed was avoided. Experimental results show that the data update rate of the vision measurement system with proposed method is up to 186.33 frame/s base on 5 markers with the resolution of 2048×2048. Comparing with the time-consuming of global search and neighborhood search without the proposed structure, the speed is increased by 33.5 times and 2.7 times, which extends the application of the system to the measurement of high-speed moving targets.