Detection of Dumpster Fill Level Using Computer Vision

The paper considers a solution to the problem of delayed garbage collection using computer vision and machine learning. It is proposed to carry out photo and video recording of the condition of container sites where dumpsters are located, after which, using neural network models, determine the amount of accumulated garbage. To solve this problem, an extensive selection of images of dumpsters of various types and shapes was prepared. Based on the ResNet architecture, a neural network has been trained to classify dumpsters. With the help of a trained Yolo model, individual dumpsters are detected in the image.