In this project we are going to count cars that pass a virtual loop in an image taken from a video camera.
The camera is a smart camera using a combination of Raspberry Pi V1.3 camera and a Grove Vision AI V2 module. This in turn is connected to a Xiao ESP32-C3 computer which is on a Seeduino Expansion Board for Xiao.
This is a demo only and includes only one dimension for the loop, the "x" or horizontal dimension. This can easily be expanded to include the "y" direction of the loop and car and for extra virtual loops in other lanes but that needs to be calibrated for each traffic scene,
The original model was developed for the Nvidia Jetson Nao and this resulted in an inference speed of 17 inferences per second. Our deployment on the Grove Vision AI V2 will not exceed this but should come close.
The firmware based on Nvida's TrafficCamNet in the Grove Vision AI V2 checks if the car is within the virtual loop. The car's boundary in the x dimension is xCarBegin on the left and xCarEnd on the right. Only cars that are within the virtual loop boundary and have a transition across the loop are counted.
For the Xiao ESP32-C3 its software was written in the Arduino IDE and libraries for the Xiao Expansion Board for the OLED display and the Grove Vision AI V2 were used.
This project has been entered into the Seeed Studio Vision AI Challenge and complements another project I have published on Seeed Studio using magnetic induction loops to count cars.
Perhaps one additional projects on traffic: using a Smart Camera as a safety device on road construction sites.
References:
* Counting Cars with Induction Loops using Seeeduino ( www.seeedstudio.com/Counting-... Seeeduino-p-5219.html?___store=retailer )
* TrafficCamNet Model Card ( catalog.ngc.nvidia.com/orgs/n... )
* Seeeduino XIAO Expansion board ( www.seeedstudio.com/Seeeduino... )
* Grove Vision AI V2 ( www.seeedstudio.com/Grove-Vis... )
* SenseCraft AI ( seeed-studio.github.io/SenseC... )
Негізгі бет Counting Cars with the Grove Vision AI V2
Пікірлер: 1