ADAPTIVE MULTI-SENSOR FUSION BASED INTELLIGENT WASTE SEGREGATION SYSTEM

Abstract

In the past few years, the quick urbanization and the use of manual sorting has resulted in the growing environmental and operational burden of MSW management and reduced the recycling efficiency and reliance on landfills. In this work, we are introducing an Intelligent garbage Segregation System with Adaptive Multi Sensor Fusion and AI in Cloud, which can autonomously separate the garbage into Metallic stream, biodegradable stream and non-biodegradable stream. The initial decision stage of the pipeline is an inductive proximity sensor which identifies metallic objects by electromagnetic induction instantaneously such that such objects can be diverted into a specialised bin without any image based calculation. The waste not detected by the inductive sensor is scanned using ESP32-CAM module and the captured frame is communicated by Wi-Fi to the backend server which is then sent to the OpenAI Vision Model for semantic classification into biodegradable or nonbiodegradable. The projected class is then sent back to the controller (ESP32) which activates a servo to move the item to the correct bin and then reset for the next class. The complete hardware-software system consisting of the inductive sensor, ESP32-CAM, servo motion mechanism and cloud based vision inference was developed and functionally tested with numerous experiments conducted on real metallic, organic and synthetic waste materials. The evaluation proved operational viability end-to-end, ensured the correct activation of the detection pathway and a sorting action as desired. The proposed cascade architecture, by limiting image-based inference to objects that actually need it, and by outsourcing the classification task to a visionlanguage model in the cloud rather than an on-board classifier, provides a low cost, scalable, and easily updatable framework, suitable for smart-city solid waste infrastructure, for automated source-level waste segregation. Keywords— Smart Waste Segregation, ESP32-CAM, OpenAI Vision Model, Inductive Proximity Sensor, MultiSensor Fusion, Embedded Systems, Internet of Things, Servo Actuation, Waste Classification.