YOLOX on Embedded Device With CCTV & TensorRT for Intelligent Multicategories Garbage Identification and Classification

Abstract

Today’s garbage classification is not fully automated and mostly relies on manual labor. In order to efficiently classify garbage and increase resource re-usability, society needs a modern solution. The development of artificial intelligence, especially in object detection, provides an excellent opportunity to develop real-time accurate garbage detection. This project is about real-time garbage detection using YOLOX and is used to detect seventeen categories of garbage. After the camera detects the garbage, it will automatically identify and analyze the type and specific the position of the garbage, which will be automatically picked up by a mechanical jaws. The datasets used in this paper consists of 5000 images with label. With the officially supported TensorRT optimization and acceleration technology, the model can be used in low-cost embedded devices and can detect garbage in near real-time without the need for a stable Internet connection. Our evaluation and comparison of these models includes some key metrics such as mean accuracy (mAP), inference time per frame, floating point operations(FLOPs), and the number of parameters of the model. The test results show that YOLOX’s mAP <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0.5:0.95</sub> can exceed 97% and the inference speed can exceed 32 fps. The performance far exceeds that of other YOLO series, with high accuracy and speed in complex environments, making it highly applicable to today’s garbage classification needs.