A STUDY ON AGGREGATE VOLUME ESTIMATION IN STOCK YARD WITH AI BASED IMAGE RECOGNITION

Abstract

 This paper attempts to estimate the aggregate volume in an outdoor asphalt plant by using a single camera and Convolution Neural Network (CNN) based-image processing. An end-to-end prediction model between a captured image and a volume was used. However, its accuracy was heavily affected from sunlight. To solve this problem, a new data augmentation technique is employed for the effective training. This technique generates virtual training images with many variations of sunlights. Our CNN trained by this dataset has suppressed the estimation error to about 8%. The experimental results have shown that the proposed method is robust to the outdoor noises.