Real-Time Image Preprocessing for Abandoned Object Detection in Urban Surveillance

In this article, we offer a real-time image preprocessing pipeline, which is intended to improve the accuracy and dependability of computer vision models, more precisely a model that identifies abandoned objects in public locations, especially in busy cities where unattended things might create safety concerns. We implement a frame sampler, a resolution changer, a noise reducer, and a blurring algorithm for privacy reasons. With videos coming from security cameras, the steps ensure the system is efficient and respectful of privacy, it's all installed on a NVIDIA Jetson Nano. We present meaningful results when it comes to processing speed thanks to the pipeline, it also helped reduce false alerts. In addition, we conducted a case study on a Moroccan university campus, showing how the system functions effectively under real situations.