Vision Based Littering Detection and Face Tagging using multi-model Architecture

In this research, a new framework has been introduced to detect illegitimate littering action using multi-model architecture based on computer vision methods. The methodology aims to address the pervasive issue of littering by leveraging computer vision techniques. It utilizes a multi-model architecture, integrating pose detection, object detection and face recognition algorithms to identify relation between joints of a person and object held by them. Also, face tagging method is performed to mark the person as potential litterer which was not previously introduced in existing literature. This study conducted multiple trials which showed effectiveness of the model in real-world instance. Results showed overall precision of 0.947 and recall of 0.857 over a significant number of trials. The experimental analysis for individual components of this model has been discussed in ablation. With its potential to significantly mitigate the impact of littering, this innovative approach holds great promise for creating cleaner and more sustainable communities.