Spatial-temporal human gesture recognition under degraded conditions using three-dimensional integral imaging: An Overview

In the field of machine learning and computer vision, recognizing a human gesture in degraded environments is a challenging issue. We overview a method that combines a three-dimensional (3D) imaging system known as integral imaging (InIm) with 3D correlation filters to recognize occluded human gestures under low illumination conditions. The integral imaging system captures multiple images, with different perspective information, which allows for the reconstruction of 3D images with both lateral and depth information. The 3D reconstructed image has a higher SNR than the captured two-dimensional (2D) image. After 3D reconstruction, the total variation (TV) algorithm is applied to denoise the 3D images. A 3D non-linear distortion-invariant filter is used for human gesture recognition. In degraded environments, we show that the proposed method is more robust for human gesture recognition compared with conventional recognition algorithms based on space-time interest points (STIP) feature detector and 3D histogram of oriented gradients (HOG) feature descriptor with support vector machine (SVM).