In the vision measurement for high speed moving target, the image sequence with high resolution and high frame rate brings a lot of datas. How to recognize the cooperation target quickly and extract the feature information in real-time is becoming a challenge in vision measurement system. In each clock cycle of high speed camera, multi- pixel is transmitted in parallel. In this situation, a hardware accelerating structure based on multi- dimensional pyramid is proposed, the task of global searching and connected components labeling is realized, the features of connected components are extracted real- timely based on the labeling results. In field programmable gate array, with the pyramid structure formed by two- dimensional processing element (PE) array and multi-label feature statistical structure formed by one-dimensional PE array, the high density computing process of data flow is distributed into each PE node equally. Combining with the parallel pipeline, the global searching and labeling process which takes a remarkable time- consuming in ordinary vision system is accomplished synchronously with image transmission. It is verified that the distributed computing structure for feature extraction can deal with real- time data processing at the flow rate of 680 Mpixel/s at the frequency of 85 MHz with CameraLink interface. As a preprocess part, it can be used in high-speed vision measurement system.