Illegal waste dumping detection

Illegal waste dumping represents a serious environmental and public health challenge, motivating the development of automated surveillance systems capable of detecting such events in real time. This paper describes and analyzes the results of the Illegal Waste Dumping Detection (IWDD) international contest, which aims to advance video-based methods for recognizing illegal disposal activities from fixed surveillance cameras. We describe the Mivia-IWDD dataset introduced for the competition, consisting of 400 video clips (200 positive, 200 negative) with precise temporal annotations, covering both static and dynamic dumping actions as well as challenging negative scenarios across diverse environmental conditions. Ten teams participated in the contest, proposing heterogeneous approaches based on spatio-temporal deep learning, action recognition, temporal modeling, and efficiency-oriented design choices. We evaluated the methods using a comprehensive protocol that combines classical detection metrics (Precision, Recall, and F1-score) with additional indicators targeting real-time applicability, including notification delay, processing frame rate, and memory usage. Moreover, we analyzed and compared the results achieved by all teams from multiple perspectives. Beyond ranking performance, this paper provides useful insights and highlights open challenges and promising research directions, contributing a benchmark and practical guidelines for future work on illegal waste dumping detection in smart surveillance systems.