LTDTS: A Lightweight Trash Detecting and Tracking System

Zijun Yu, Jin Liu, Xingye Li

Abstract: Community environment is one of the focuses of urban governance. And the phenomenon of residents littering in violation of community regulations is one of the most prominent problems in the management of the community environment. However, the existing methods, whether it is time-consuming and labor-intensive human management or action detection-based system that still cannot achieve a good balance between accuracy and real-time performance, cannot solve this outstanding problem well. At the same time, the more commonly used object detection-based systems also have poor performance due to the complexity of the model and system architecture. To address these issues, we propose a novel lightweight trash detection system. The system uses an improved yolov5 algorithm, which is more suitable for the detection of small targets like litter. In addition, we novelly proposed two methods called tracking object transmission and video backtracking, combined with the tracking algorithm based on kernelized correlation filter, we successfully achieved accurate localization of littering pedestrians. So far, our model has been experimented on several small public datasets and our self-designed community littering dataset. What's more, our system has also been put into use in some communities, with initial results that are acceptable, which successfully verified the feasibility of our method.