Intelligent Sparse Visual Motion Estimation Method Using Local Activation Model for Multiaperture Imaging Systems

Adaptive visual motion perception has major engineering significance in bionic vision for Internet of Things (IoT) applications, particularly for autonomous motion platforms. However, the most advanced multiaperture array imaging systems (MAS), calculating global dense optical flow is often computationally prohibitive and conflicts with the motion-sensitive characteristics of biomimetic compound-eye (BCE). Here, we report a motion estimation method that joint sparse optical flow estimation by selectively activating aperture motion flow to minimize redundant calculations. Furthermore, to solve the aperture correlation matching challenge, a scale-invariant feature transform (SIFT) acceleration network is proposed based on the customized aperture descriptors, employing a multilayer perceptron (MLP) for feature separation. Finally, within the framework of graph signal process (GSP) and sparse coding, our results validate the feasibility of the proposed strategy, offering a promising solution for visual motion measurement in intelligent bionic vision systems.