Browsing by Author "Degardin, Bruno Manuel"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- Weakly and Partially Supervised Learning Frameworks for Anomaly DetectionPublication . Degardin, Bruno Manuel; Proença, Hugo Pedro Martins CarriçoThe automatic detection of abnormal events in surveillance footage is still a concern of the research community. Since protection is the primary purpose of installing video surveillance systems, the monitoring capability to keep public safety, and its rapid response to satisfy this purpose, is a significant challenge even for humans. Nowadays, human capacity has not kept pace with the increased use of surveillance systems, requiring much supervision to identify unusual events that could put any person or company at risk, without ignoring the fact that there is a substantial waste of labor and time due to the extremely low likelihood of occurring anomalous events compared to normal ones. Consequently, the need for an automatic detection algorithm of abnormal events has become crucial in video surveillance. Even being in the scope of various research works published in the last decade, the state-of-the-art performance is still unsatisfactory and far below the required for an effective deployment of this kind of technology in fully unconstrained scenarios. Nevertheless, despite all the research done in this area, the automatic detection of abnormal events remains a challenge for many reasons. Starting by environmental diversity, the complexity of movements resemblance in different actions, crowded scenarios, and taking into account all possible standard patterns to define a normal action is undoubtedly difficult or impossible. Despite the difficulty of solving these problems, the substantive problem lies in obtaining sufficient amounts of labeled abnormal samples, which concerning computer vision algorithms, is fundamental. More importantly, obtaining an extensive set of different videos that satisfy the previously mentioned conditions is not a simple task. In addition to its effort and time-consuming, defining the boundary between normal and abnormal actions is usually unclear. Henceforward, in this work, the main objective is to provide several solutions to the problems mentioned above, by focusing on analyzing previous state-of-the-art methods and presenting an extensive overview to clarify the concepts employed on capturing normal and abnormal patterns. Also, by exploring different strategies, we were able to develop new approaches that consistently advance the state-of-the-art performance. Moreover, we announce the availability of a new large-scale first of its kind dataset fully annotated at the frame level, concerning a specific anomaly detection event with a wide diversity in fighting scenarios, that can be freely used by the research community. Along with this document with the purpose of requiring minimal supervision, two different proposals are described; the first method employs the recent technique of self-supervised learning to avoid the laborious task of annotation, where the training set is autonomously labeled using an iterative learning framework composed of two independent experts that feed data to each other through a Bayesian framework. The second proposal explores a new method to learn an anomaly ranking model in the multiple instance learning paradigm by leveraging weakly labeled videos, where the training labels are done at the video-level. The experiments were conducted in several well-known datasets, and our solutions solidly outperform the state-of-the-art. Additionally, as a proof-of-concept system, we also present the results of collected real-world simulations in different environments to perform a field test of our learned models.