UMAD: University of Macau Anomaly Detection Benchmark Dataset

TL;DR

UMAD introduces the first anomaly detection with reference benchmark tailored for robotic patrolling, enabling anomaly localization via semantic changes between aligned reference and query frames. It proposes an adaptive image warping method to achieve near-pixel alignment and evaluates state-of-the-art scene change detection models in both binary and multi-class ADr settings. The dataset comprises six outdoor scenes, 120 sequences, 26k image pairs and 140k labels across seven semantic classes, with ground-truth annotations focused on ground-plane changes. Together, the dataset, alignment technique, and baseline evaluations establish a foundation for developing practical ADr methods in real-world patrol scenarios.

Abstract

Anomaly detection is critical in surveillance systems and patrol robots by identifying anomalous regions in images for early warning. Depending on whether reference data are utilized, anomaly detection can be categorized into anomaly detection with reference and anomaly detection without reference. Currently, anomaly detection without reference, which is closely related to out-of-distribution (OoD) object detection, struggles with learning anomalous patterns due to the difficulty of collecting sufficiently large and diverse anomaly datasets with the inherent rarity and novelty of anomalies. Alternatively, anomaly detection with reference employs the scheme of change detection to identify anomalies by comparing semantic changes between a reference image and a query one. However, there are very few ADr works due to the scarcity of public datasets in this domain. In this paper, we aim to address this gap by introducing the UMAD Benchmark Dataset. To our best knowledge, this is the first benchmark dataset designed specifically for anomaly detection with reference in robotic patrolling scenarios, e.g., where an autonomous robot is employed to detect anomalous objects by comparing a reference and a query video sequences. The reference sequences can be taken by the robot along a specified route when there are no anomalous objects in the scene. The query sequences are captured online by the robot when it is patrolling in the same scene following the same route. Our benchmark dataset is elaborated such that each query image can find a corresponding reference based on accurate robot localization along the same route in the prebuilt 3D map, with which the reference and query images can be geometrically aligned using adaptive warping. Besides the proposed benchmark dataset, we evaluate the baseline models of ADr on this dataset.

Figures (7)

• Figure 1: Overview of data collection and data pre-processing.
• Figure 2: Examples of anomalous objects in the UMAD dataset.
• Figure 3: Flowchart of Adaptive Warping. Img1 and Img2 represent a pair of reference and query images.
• Figure 4: Comparison of our scores and COLMAP scores under different quantities and distributions of inliers in the image for L = 4. In the first row, with 16 points, the ratio of our maximum score to the minimum score is four times. This indicates that our score exhibits a higher level of discrimination. In the second row, compared to the scenario with 16 points concentrated on the left side, our score tends to favor a more evenly distributed scenario with 12 points on the right side.
• Figure 5: The translation statistics between the query images and reference images in a sequence of 5,556 raw data image pairs from Scene 1, Sequence 00.