Single-Pixel Estimation for Enhanced Object Detection and Recognition with Kinect for Windows V2

Microsoft Kinect sensor has been used for all sort of computer vision projects since inception, as it provides us with both RGB and depth data. With a high range of depth information provided by the Microsoft Kinect v2 Infrared (IR) depth sensors, there has been increased attention in utilizing this depth information for detection and tracking. In this paper, we proposed a Single-pixel-grid based method for calculating the pixel of an object that is the closest or highest within a particular threshold. After establishing the record holding pixel object, we developed an algorithm for detecting and tracking the location of the object based on the pixel. At the end of the experiment, results show that using this algorithm, the Kinect v2 was able to detect the pixel that is the closest or highest in two tested thresholds and as well tracked accurately the object with the record pixel point. Analysis and comparison of results shows improved accuracy in object location detection using our algorithm.