Internet-Of-Things People Counter
Team Members Heading link
- Jiaxing Chen
- Dongyu Li
- Hongbin Sun
- Yi Zhang
Advisor: Vladimir Goncharoff, PhD
Project Description Heading link
Our team sought to design a video-based system that monitors the number of people in large and medium-sized places to assist managers and security teams in redistributing human resources. The need for such a system exists in most retail stores. For example, queues of people often form at service counters and point-of-sale terminals, waiting to be served. Management would like to respond quickly by sending more employees to service the waiting customers. The system that we designed automatically analyzes a video camera signal, detects and counts the number of people in real-time, and alerts a central computer when a crowd exceeding a specified number has gathered. Our system is designed to have many such units, with each unit doing its own image signal processing so as to be easily expandable. The communications between each camera node and the central monitoring node is according to the Internet-Of-Things model, where each camera node is accessed via WiFi. The most challenging aspect of our project was to detect the number of people seen in an image: due to inherent attributes of biological characteristics, human beings generally look the same but on closer inspection have significant individual differences. We found that detecting faces is the easiest way to identify the presence of people in an image. For this task, we used a Multi-Task Cascaded Convolutional Neural Network (MTCNN) composed of a PNet (Proposal Network), RNet (Refinement Network), and ONet (Output Network) for face detection and human detection. The neural network algorithms will easily run in Python programming language on a dedicated Raspberry Pi computer.