Comparison of Implementation of Computation of Video Image Displacement Vectors in Built-in Systems

Abstract

Various algorithms of video image displacement vectors calculation are considered. These algorithms are used in smartphones, digital photo cameras, different special-purpose devices in order to stabilize the image, traffic control, air photographic survey, object tracking and other applications. In order to compare the efficiency of hardware resources consumption to perform video images displacement vectors calculation algorithms there were considered: Lucas-Kanade method-based algorithm and the image matched filtering algorithm. Global image displacement was estimated by means of a hardware platform – Ultrascale+ Xilinx SoC. Matrix calculations of the Lucas-Kanade algorithm were performed with four equal hardware blocks, and it made it possible to calculate the displacements of the frames for four points at the same time. The acquired average optical flow calculation time was 7.5 ms. Four hardware blocks were also used to perform parallel calculations to implement the 2D-DFT-based image matched filtering. The resulting operation time of the algorithm was 2.9 ms, and the CPU time used by the algorithm was 0.2 ms. The implementations revealed weak spots of the algorithms, and the required hardware resources of SoPC were determined. The matching filtering algorithm proved to be more efficient as compared to the Lucas-Kanade algorithm to estimate the video sequence consecutive frames displacement upon enough hardware resources.