Overview
This term, we've been working hard to build a people counting app. The use case of this application is to place a camera on overhead entrance doorways like the entrances to Walmart, Costco, grocery stores, theaters, malls, and even the streets. The camera would count how many people would come through and record this. The data could be seen graphically through an Android interface. The result could inform users how many people come through at different times in the day for purposes such as determining the peak customer count, the rate of customers coming in, and other generic analytics.
Demo
How it works
The diagram below illustrates the full workflow of the people counter app. A ZED 2 stereo camera performs person tracking on a Jetson TX2 Linux machine.
Hardware:
The number of people seen by the camera would be aggregated every minute and then sent up to the Redis time series database for storage. Additionally the hardware would also stream video data over the internet via AWS Kinesis Video Streams. For remote monitoring, configuration, debugging, and updates, AWS IoT Greengrass was deployed on the hardware. It allows for remote SSH tunneling, deployment of programs such as Python and Docker containers, and MQTT messaging.
Frontend / Backend:
On the Android side, the user would be able to view the live video, past recorded video, and a graphical interface of how many people have been seen by the camera. The graph can display the number of people within a selected time range and a specified granularity (e.g. minutely, hourly, daily, monthly, yearly). The graph communicates with a local backend Flask API that serves the time series data from Redis. For the video streaming, a Java Kinesis client was implemented that performs video playback using an HLS player.
MVP Winner
We want to give a special shoutout to Alex Munsee who was our MVP winner for the term, winning $100. He worked on the graph interface that displayed the count of the number of people seen by the camera.
Comments