Search

We present MultEYE, a traffic monitoring system that can detect, track, and estimate the velocity of vehicles in a sequence of aerial images. The presented solution has been optimized to execute these tasks in real-time on an embedded computer installed on an Unmanned Aerial Vehicle (UAV). In order to overcome the limitation of existing object detection architectures related to accuracy and computational overhead, a multi-task learning methodology was employed by adding a segmentation head to an object detector backbone resulting in the MultEYE object detection architecture. On a custom dataset, it achieved 4.8% higher mean Average Precision (mAP) score, while being 91.4% faster than the state-of-the-art model and while being able to generalize to different real-world traffic scenes. Dedicated object tracking and speed estimation algorithms have been then optimized to track reliably objects from an UAV with limited computational effort. Different strategies to combine object detection, tracking, and speed estimation are discussed, too. From our experiments, the optimized detector runs at an average frame-rate of up to 29 frames per second (FPS) on frame resolution 512 × 320 on a Nvidia Xavier NX board, while the optimally combined detector, tracker and speed estimator pipeline achieves speeds of up to 33 FPS on an image of resolution 3072 × 1728. To our knowledge, the MultEYE system is one of the first traffic monitoring systems that was specifically designed and optimized for an UAV platform under real-world constraints. [ABSTRACT FROM AUTHOR]

We present an unsupervised deep learning approach for post-disaster building damage detection that can transfer to different typologies of damage or geographical locations. Previous advances in this direction were limited by insufficient qualitative training data. We propose to use a state-of-the-art Anomaly Detecting Generative Adversarial Network (ADGAN) because it only requires pre-event imagery of buildings in their undamaged state. This approach aids the post-disaster response phase because the model can be developed in the pre-event phase and rapidly deployed in the post-event phase. We used the xBD dataset, containing pre- and post- event satellite imagery of several disaster-types, and a custom made Unmanned Aerial Vehicle (UAV) dataset, containing post-earthquake imagery. Results showed that models trained on UAV-imagery were capable of detecting earthquake-induced damage. The best performing model for European locations obtained a recall, precision and F1-score of 0.59, 0.97 and 0.74, respectively. Models trained on satellite imagery were capable of detecting damage on the condition that the training dataset was void of vegetation and shadows. In this manner, the best performing model for (wild)fire events yielded a recall, precision and F1-score of 0.78, 0.99 and 0.87, respectively. Compared to other supervised and/or multi-epoch approaches, our results are encouraging. Moreover, in addition to image classifications, we show how contextual information can be used to create detailed damage maps without the need of a dedicated multi-task deep learning framework. Finally, we formulate practical guidelines to apply this single-epoch and unsupervised method to real-world applications. [ABSTRACT FROM AUTHOR]