| Optical flow is the apparent motion of the brightness mode of moving image, includingthe information of the moving object, it can be used to indicate the movement variation andthe movement situation of the determination target. The use of optical flow field candistinguish different moving objectsThe changes in the expression of motion in picture. Therefore, it can be used todetermine the movement of the target by viewers. It can used to distinguish different movingobjects and characterize the direction and speed of movement of objects in order to obtainbackground and foreground movement to conduct a series of analyzes and applications.Optical flow field can be divided into local sparse optical flow calculation and globaldense optical flow calculation. With the image/video resolution continues beyond HD, FHD,UHD,4K level, optical flow computation based on traditional computing platforms has beenfar unable to meet the practical applications’ real-time requirements. So using hardwareacceleration to achieve high computational performance low power consumption has becomea hot issue in this field.This paper focuses on optical flow calculation method based on FPGAs. First, thealgorithm theory of existing algorithms of spares and dense optical flow field is elaboratedand two types of optical flow algorithms’ parallelism and commonality is analyzed. Second,we make standardized hardware design to each core computing processes in the form ofcommon IP core, and make corresponding verification and packaging. And then, we makecombining optimization by system-level design to different IP cores, thus, achieve two typesof optical flow calculation hardware systems. Finally, these two systems are merged into ahardware accelerating system with high computation performance and versatility. This systemis able to achieve high-precision test sets320x240dense optical flow calculation40Fps andsparse optical flow calculation1390Fps, which can meet the demands of applications in realtime. |
PDF Full Text Request