Nowadays, high-speed video microscopy is used in many applications like microrheology1, 2 or flow cytometry3 to measure mechanical properties of cells or to identify their type. Typically, high-speed cameras use buffering to reach very high frame rates due to the limited bandwidth of the interface to a PC like Ethernet or USB. Additionally, analysis of large data is compute-intense and in many cases difficult to do online. We developed a system that consists of a high speed CMOS image sensor combined with a field programmable gate array (FPGA) and a pulsed LED illumination system. Due to an image transformation that is done on the FPGA, the dimensionality of the data is reduced without loss of important information. This leads to a significant reduction of the amount of data as well as to noise reduction as a side effect. Furthermore, we developed a modular analysis toolkit that can be used to do the whole analysis directly on the same FPGA online so that buffering is not required and measurements can run continuously on high frame rates. Hence, we can analyze a large total number of objects at very high throughput rates in microfluidic devices. We present the analysis of diluted whole blood in a microfluidic system with our device as well as a sorting application that uses multiple regions of interest that are observed simultaneously so that particles can be analyzed before and after a manipulation or gate.