High-speed vision system that can acquire and process images or videos at a high rate (hundreds or thousands frames per second) has significant advantages over standard video cameras operating at dozens of frames per second in areas of tracking, robot control, industrial manufacturing, navigation and motion analysis and so on. The main challenge to achieving high-speed performance is how to manage and coordinate image processing and system controlling in an extremely short period of time. In this talk, our developed high-speed vision systems with intelligent tracking are introduced including: 1) high-speed multithread active vision with ultrafast mirror-drive devices, 2) super-telephoto detecting and tracking with deep learning, and 3) vibration source tracking with pixel-level high-speed vision signal processing. These proposed systems show how high-speed vision can combine ultrafast optomechatronic devices and deep intelligent processing for real-world applications such as high-speed targets detecting, tracking and zooming at a distance.