The proliferation of closed-circuit television cameras in urban environments has engendered a vast repository of visual data, presenting an opportunity to combat the pervasive issue of littering. This paper presents a novel approach to real-time detection and monitoring of littering using computer vision techniques, specifically OpenCV2 and lightweight YOLO V3 object detection using COCO names which contain labels of images, whose performance is enhanced by the efficiency of TFLite. The research presented herein contributes to the advancement of urban maintenance and cleanliness, underscoring the potential of smart city initiatives in leveraging YOLO’s streamlined object detection and the optimized performance of TFLite contributes significantly to the algorithm’s efficiency. The integration of these technologies in this context serves as a promising case study for the practical application of cutting-edge computer vision techniques in addressing this societal challenge. Neural network variants are a popular choice for object detection. The proposed approach efficiently classifies the litter using the YOLO V3 model, a real-time object detection system that excels in both accuracy and speed. It differs from traditional object detection methods by dividing the task into a single neural network that predicts bounding boxes and class probabilities directly from the full image in one forward pass. Our study incorporates a specifically crafted dataset, accessible through the provided drive link, where we have created a 4-h-long diverse dataset with 23 instances of littering with timestamps against different backgrounds serving as the testing ground for our real-time littering detection system. The central logic of our methodology hinges on the temporal evolution of signals obtained from video frames over time. By capitalizing on time series signal processing, our approach effectively discerns the nuanced patterns and characteristics distinguishing litter objects from human hands within the video data. This model can be used to collect data for deep learning models. We use an analytical computer vision approach due to the lack of a specific dataset of littering videos. Further, our algorithm uses YOLO as a fundamental building block, bestowing us with lightweight and rapid detection compared to heavy video-based models. The proposed approach provides a robust mechanism for identifying and recording instances of littering in a real-time video feed.