Crowd Counting and Intelligent Warning System based on Raspberry Pi and Neural Compute Stick 2
- 0 Collaborators
With the rapid increase of urban population and surveillance cameras, crowd scene analysis is of great importance. We are developing an application system to perform crowd counting and crowd scene analysis on the edge computing device. ...learn more
Project status: Under Development
Intel Technologies
OpenVINO,
Movidius NCS
Overview / Usage
With the rapid increase of urban population, crowd scene analysis has attracted more and more attention. We are developing an edge computing device for crowd counting and safety pre-warning. This system is based on Raspberry Pi and Neural Compute Stick 2, and can be deployed on any existing surveillance camera to perform intelligent scene analysis.
Methodology / Approach
Our core algorithm for crowd counting is based on convolution neural networks. Considering people are always occluded in crowd scenes, we utilize the density-map based method to regress the crowd density map and figure out the number of people. The alarming function performs on the generated density map to recognize crowd regions with high density or high change rate of density. To deploy the algorithm in real application scenarios, we bulid an edge computing environment with Raspberry Pi 4 and Neural Compute Stick2 (NCS2). Specifically, NCS2 serves to load deep models and perform efficient inference after optimization, while Raspberry Pi 4 handles the resource scheduling and the remaining computing tasks.
Technologies Used
Convolutional Neural Networks
Raspberry Pi 4
Neural Compute Stick2 (NCS2)
Documents and Presentations
Repository
https://github.com/yongtuoliu/Crowd-Counting-via-Cross-stage-Refinement-Networks