Detection of waste containers using computer vision

This work is a part of an ongoing study to substitute the identification of waste containers via radio-frequency identification. The purpose of this paper is to propose a method of identification based on computer vision that performs detection using images, video, or real-time video capture to identify different types of waste containers. Compared to the current method of identification, this approach is more agile and does not require as many resources. Two approaches are employed, one using feature detectors/descriptors and other using convolutional neural networks. The former used a vector of locally aggregated descriptors (VLAD); however, it failed to accomplish what was desired. The latter used you only look once (YOLO), a convolutional neural network, and reached an accuracy in the range of 90%, meaning that it correctly identified and classified 90% of the pictures used on the test set.