Real-time optical flow measurement based on GPU architecture

In this paper we propose an optical flow (a velocity field of the appearance in the image) estimation method using Graphics Processing Unit (GPU). To estimate optical flow, we use Spatiotemporal Differentiation Method^[1]. In order to improve the feature that this method is influenced easily by the noise in the image, we use Multichannel Integration Method^[3]. However, real-time estimation of the optical flow with due accuracy is not yet realized. So, we focus on the most time-consuming part in the program is repetition of simple operation, and implement GPGPU architecture this program. Experimental results show that the processing time is reduced except first processing.