A prototype of video detection system applied to traffic flow inspection is developed, which uses CMOS linear image sensor with high resolution 2K pixels and wide dynamic range as the core of imaging device. It combines FPGA with DSP as the core of acquisition and processing of massive image data. Moreover, a novel multiscale and hierarchical clustering algorithm for image segmentation is presented. Based on the theory of graph spectral, the algorithm can construct a new graph by analyzing the feature of an original image at different clustering scales, so that image segmentation can be accomplished easily to segment the image. The simulation results show that the row scan speed of this system can reach to 1000 lines per second, the resolution being 2048 pixels.