An Accurate Crowd Counting App

Abstract

In the past few years, the argument regarding the crowd counting of march and rally has captured society's attention. Many parties that participated in the rally crowd counting are releasing different crowd counting result non-scientifically. It is difficult to recognize the reality of public opinion throughout society with the inaccurate result. This project aims to create a free iOS app where users can count on the rally accurately. User can perform gestures on the video for annotating and counting. By using this app, it can significantly reduce the workforce on-site and increasing efficiency. The development of this iOS app is through the platform of X-code with Swift programming. In this app, user can manually annotate the people inside the video. Hence, the vision framework is the backbone where performing computer vision algorithm on detecting multiple objects for the input video. While using the API from the vision framework and the designed counting algorithm, the annotated images in the video can be tracked and counted. User can now manually quote the image of people and trace it. Also, they can adjust the annotation size by using the stepper. User can create counting line by themselves. When the annotated frame has passed the line, the system will count. Machine learning model should be adopted to perform accurate object detection and counting, where the model can be created through using Create-ML from X-code. This improvement should include the Core-ML framework in the project so that the system can detect human preciously.