Intelligent Waste Sorting Platform Based on Taurus & Pegasus
Xiang Chen, Liguo Tian, Meng Li, Jinqi Liu, Yuesong Wang, Donghui Fang · 2022
Garbage sorting is an action for everyone to participate, but the existing garbage sorting mainly relies on manual discrimination. In this paper, an intelligent garbage sorting platform is designed using Taurus & Pegasus kit, which records the video stream of garbage using the sensor side of Taurus vision development board. The generated video files are then copied to an SD card for dataset creation, data cleaning, and data annotation. ResNetl8 network model is built using Pytorch framework for training (get pytorch), model transformation (model-caffe), quantization and simulation. Do end-to-end modeling and visualization. Network models are combined with HiSilicon NNIE deep learning acceleration engine to implement fast target classification algorithms. Finally, the model is deployed to the Taurus side for image recognition using an SD card. Residents place garbage in front of the camera, and the Taurus side recognizes and displays the garbage classification inference results and probabilities on the screen. At the same time, the result is sent to Pegasus via serial communication. Pegasus receives the classification result and drives the corresponding steering engine to open the garbage can lid and give voice prompt. When no object or background picture is captured in the camera screen, the trash can lid will be closed.