Smart Illegal Dumping Detection

摘要

Illegal dumping has been a chronicle problem in many cities in the world. The odors and contaminants caused by abandoned household items and dumped garbage, and construction leftovers not only ruin the city view but also threaten citizens health. To reduce the illegal dumping, a few cities have designed community-based voluntary reporting systems and surveillance-camera-based monitoring systems. However, these approaches still require manual monitoring and detection, which are costly and vulnerable to false alarms. In this paper, we propose to use deep learning approach to recognize various types of frequently dumped wastes. To achieve higher accuracy, we explored various approaches and demonstrate the accuracy variance with regard to the number of classes, baseline models, and input image characteristics. We also propose to use edge computing to reduce unnecessary image transfer to the servers. The edge computing station runs a deep learning model for captured images of individual dumping hot spots and sends the images to the server only when the image contains frequently dumped wastes. To successfully deploy deep learning models to edge computing stations that are shipped with limited resources, we also apply the state-of-the-art deep learning model compression tool. Our experimental results show that the proposed approaches provide high recognition accuracy with small memory footprint.