Real-time method for counting unseen stacked objects in mobile

A novel method that automatically counts number of objects piled or stacked to each other in mobile device image is proposed. Distinct from other methods, the proposed method automatically determines the size and shape and detects the template of unknown to-be-counted objects. Such template is then used to perform fast template matching all over the input image to count the number of the same objects. The similarity function for determining the detected object has been proposed. The evaluation on speed and accuracy was performed by testing on images of various construction materials captured from on real site by the Android smart phone. The result suggests that the proposed method outperforms all baselines including the recent lattice detection based approach. The computation time is sufficiently fast for real-time use in mobile devices. It can also efficiently deal with the problem of information loss (i.e., lost edges) due to various unpredicted lighting condition. In term of time, the comparison of counting time was also done between the method and human.