AUTOMATIC PLASTIC BOTTLE SORTING WITH ROBOTIC ARM USING RASPBERRY PI

Abstract

Plastic pollution is a major threat to the environment as it contaminates water bodies, soil and air. Plastic bottles are among the most common pollutants since they are often used but are disposed improperly. A significant number of recyclable items wind up in landfills or pollute natural environments due to traditional sorting techniques. This inefficiency increases pollution and harms ecosystems. This project tackles the pressing issue of plastic waste management by developing an automated system that integrates computer vision with a custom 6-DOF robotic arm, powered by a Raspberry Pi. The system employs a Single Shot Detector (SSD) MobileNet model, trained on locally sourced images, to detect and sort plastic bottles in real-time. The robotic arm, constructed with 3D-printed components, executes the sorting tasks with a 90% success rate. The software, written in Python, combines image processing, machine learning, and hardware control, achieving an average precision of 81.1% at IoU 0.50 and detecting bottles with confidence levels between 80% and 95% across various environments. While the project demonstrates significant potential in automating waste management, it also encountered challenges. Issues such as limited datasets, model training complexities, and the Raspberry Pi's hardware constraints affected the system's performance in terms of precision, frame rate, and responsiveness. Additionally, techniques like model quantization and pruning should be explored to increase the system's efficiency and adaptability in different environmental conditions.