Real-Time Diffuse Behavior Detection of Pixels from Outdoor Image Sequence

Abstract

We have proposed a technique to detect diffuse reflectance behavior pixels in real-time from a time-lapsed low dynamic range image sequence of the outdoor scene under few assumptions, unknown dynamic illumination and without using calibration objects. Each pixel analysis is done at the arrival of the every input image of the image sequence. In this process we incrementally compute new mean and standard deviation of each pixel on the arrival of the new input image and perform Z-test on each pixel for its outlier value. Shadow and specular candidature failure of a pixel along with its inbound Z-test result signifies pixel is having diffuse behavior.