Abstract
This project introduces an intelligent real-time surveillance system designed to automatically detect and penalize individuals who drop or throw objects specifically bags, cups, and bottles below a defined safety boundary using advanced computer vision and automated identity verification. The system integrates YOLOv5 object detection with a custom lightweight tracking mechanism to continuously monitor the spatial trajectory of detected objects and identify when any of them cross a predefined ground-level threshold. Alongside object monitoring, the system employs LBPH face recognition, enabling robust identification of registered individuals even under varying illumination, partial occlusions, or moderate pose variations. Each user undergoes face enrollment through the camera, generating a localized dataset used to train an LBPH classifier stored in a persistent face model. During live surveillance, the system associates recognized users with real-time actions using a short-term face-memory mechanism that ensures stability even if the subject briefly exits the frame. The YOLO engine processes incoming frames to identify objects such as bags, cups, and bottles with high accuracy and low latency, while a bounding-box tracking layer computes object centroids and assesses rule violations. When a violation occurs—such as intentionally discarding a bag, cup, or bottle below ground level—the system logs the event and automatically triggers an SMS alert through the Twilio cloud messaging API, sending the user’s name, detected object type, and applied fine directly to their registered mobile number. A secure Flask-based web interface provides user authentication, face training, and a live video streaming dashboard, enabling administrators to monitor events, track violations, and supervise system operations in real time. All data related to users, faces, and activity events is maintained through an SQLite database for simplicity, reliability, and portability. The combined pipeline outputs a fully automated, human-independent enforcement system applicable in campuses, industrial premises, public facilities, and smart-city environments. By blending face recognition, object analytics, tracking, and cloud messaging, this project delivers a scalable, efficient, and high-precision solution for preventing littering behaviors and ensuring responsible public conduct.

 
INTRODUCTION
The rapid growth of artificial intelligence and embedded computer vision technologies has significantly transformed modern surveillance systems, shifting them from basic video monitoring solutions into intelligent, context-aware, and autonomous decision-making platforms. Traditional CCTV systems merely record video footage, placing the burden of interpretation, monitoring, and response entirely on human operators, which often leads to delays, oversight, and inefficiencies, especially in high-traffic or highly dynamic environments. To address these limitations, modern surveillance applications increasingly incorporate machine learning, real-time object detection, face recognition, automated tracking, and cloud-based alert systems. This project is an advanced implementation of such next-generation surveillance, designed to detect and prevent undesirable behaviors—specifically the act of throwing or dropping objects such as bags, cups, and bottles—within monitored areas. The system integrates real-time object detection (YOLOv5), robust face recognition (LBPH), lightweight object tracking, automated violation analysis, and Twilio-based SMS alerts, all deployed through an interactive Flask web-based interface. The motivation behind this work emerges from growing concerns over public safety, waste mismanagement, and enforcement inefficiencies in environments such as college campuses, industrial workplaces, public pathways, commercial buildings, transportation terminals, and restricted premises. Despite the availability of surveillance cameras, violations like littering, unauthorized object disposal, or harmful actions often go unnoticed due to the lack of automated monitoring, human fatigue, and the sheer scale of areas under observation. Therefore, a fully automated, intelligent, real-time monitoring solution becomes essential for ensuring cleanliness, discipline, accountability, and safety across public and private environments. The rise of deep learning frameworks has revolutionized object detection, enabling models such as YOLOv5 to recognize multiple classes—including bags, cups, and bottles—with high frame rates and impressive accuracy, even on modest computational hardware. YOLOv5’s one-stage detection architecture and optimized inference pipeline make it particularly suitable for real-time applications where rapid response is crucial. In this project, YOLOv5 is employed to detect specified objects continuously while a bounding-box tracker monitors their motion and determines whether an item has been thrown below a defined ground-threshold line. This threshold-based rule serves as the behavioral logic for identifying violations, allowing the system to classify actions not simply by detecting an object but by interpreting its motion relative to the environment. In parallel, the system incorporates an identity-verification module using the Local Binary Patterns Histograms (LBPH) method, a classical but extremely reliable approach for face recognition under varied lighting and real-time constraints. Users must first undergo face enrollment, where multiple samples of their facial images are captured and stored. This dataset is used to train an LBPH classifier, enabling precise identification of individuals appearing in the camera feed. The face-recognition module includes a temporal memory buffer, allowing the system to maintain user identity for several seconds even if the face is momentarily occluded, blurred, or out of frame. This ensures that any detected violation is accurately linked to the responsible individual without ambiguity or manual intervention. To transform the raw detection and recognition capabilities into a practical enforcement system, this project integrates a cloud communication layer using the Twilio API, enabling instant SMS notifications whenever a violation occurs. If the system recognizes a registered user and detects an object being thrown beyond the ground level, an automated alert is sent to the user’s phone number, containing the name, object detected, and penalty details. This ensures immediate accountability while creating a strong deterrent effect against unwanted actions. The entire solution is wrapped within a Flask framework, providing user registration, login authentication, live video streaming, face training, and surveillance dashboards through a browser-friendly interface. SQLite serves as the lightweight relational database backend, handling user records, face data references, and event logs with reliability and minimal overhead. The project emphasizes modularity, scalability, and ease of deployment: it can operate on standard computing hardware with a single camera and can be extended to support multiple cameras, additional object classes, or advanced behavioral rules. Furthermore, the architecture allows integration with cloud storage, centralized monitoring systems, administrative reporting dashboards, and automated fine-management systems, making it adaptable to future expansions. Overall, this project demonstrates a comprehensive integration of artificial intelligence, real-time computer vision, object analytics, identity verification, and automated communication to create a fully autonomous smart surveillance system. It transcends conventional monitoring by not only detecting objects but also analyzing human behavior, identifying offenders, and issuing automated responses without manual oversight. Such capabilities position the system as a powerful tool for smart-city development, campus discipline management, industrial safety enforcement, and public-space monitoring.

OBJECTIVES
The primary objective of this project is to develop a fully autonomous, intelligent, and real-time surveillance solution capable of accurately detecting littering activities particularly the act of throwing or dropping bags, cups, and bottles while simultaneously identifying the responsible individual and issuing immediate alerts without human involvement. The system aims to bridge the gap between traditional passive CCTV monitoring and modern AI-driven automated enforcement by integrating state-of-the-art object detection, identity recognition, behavior analysis, and communication technologies into a unified, efficient, and user-friendly platform. A key objective is the deployment of YOLOv5 as the backbone for high-speed, high-accuracy detection of targeted objects in dynamic video streams, ensuring robust operation under real-world variations such as illumination changes, background clutter, and diverse camera perspectives. The model must consistently differentiate between relevant object classes and non-target elements, thereby reducing false positives and ensuring reliable violation detection. In addition to object detection, an equally important objective is to implement LBPH-based face recognition to authenticate users in real time, allowing the system to map each detected violation to a specific individual with high confidence. This includes building a scalable face-training mechanism that captures multiple facial samples per user, optimizes the LBPH model for accuracy, and maintains a well-organized face dataset that supports incremental updates as new users are onboarded.
Another major objective is to design a lightweight tracking and behavioral analysis module capable of observing the motion path of detected objects and identifying when an item crosses below a predefined ground threshold. This objective addresses the need for interpreting not just object presence but object behavior, enabling the system to accurately classify the act of “throwing” or “dropping” based on spatial relationships and movement patterns. To enhance accountability, the system aims to maintain a short-term identity retention mechanism (face memory), allowing the system to preserve user identity even when the face briefly disappears due to occlusion or camera angle changes. Beyond detection and decision-making, a critical operational objective is to implement an automated SMS notification pipeline using the Twilio cloud messaging platform, ensuring that offenders or administrators receive immediate alerts specifying the user’s name, detected object type, and applicable fine. This automation eliminates manual intervention, accelerates response times, and enables scalable enforcement in large environments. From a software engineering perspective, the project aims to create a secure, modular, and intuitive Flask-based web interface that supports user registration, login authentication, face training, and real-time video streaming, providing administrators with full visibility into system operations. Ensuring database integrity is another objective, achieved through an SQLite backend that reliably stores user credentials, face records, and event logs while maintaining efficiency and portability.
Overall, the project’s objectives converge on building a highly responsive, accurate, and scalable surveillance ecosystem that enhances environmental responsibility, reduces manual monitoring burden, Enables automatic identification of offenders, and establishes a technology-driven disciplinary model suitable for campuses, institutions, industries, and public environments. By combining computer vision, identity analytics, event classification, and automated communication, the project seeks to demonstrate how AI-powered surveillance can deliver transformative improvements in cleanliness, safety, rule enforcement, and operational efficiency across modern infrastructures.