Efficiently locating objects and tracking them across time is a fundamental task in autonomous agents navigation. We explore different directions both in 2D and 3D. For 2D tasks, we study approaches to harness temporal prior essential for tracking, thus, performing joint detection and tracking while striking competitive speed-accuracy trade-off. While for 3D perception, where annotating objects with 3D cuboids is a costly and time consuming process, our aim is to investigate self-supervised pipelines, thus, unlocking the potential of widely-available raw data.

References

Mouawad I, Odone F. FasterVideo: Efficient Online Joint Object Detection And Tracking. To appear in ICIAPSpringer.

Mouawad I, Brasch N, Manhardt F, Tombari F, Odone F. Time-to-Label: Temporal Consistency for Self-Supervised Monocular 3D Object Detection. arXiv preprint arXiv:2203.2022

Collaboration with

Collaboration with TUM

As progress in computer vision delivers larger and more complex deep learning architectures, we are dealing with the challenges of limited resources typical of real world scenarios. Specifically, we design efficient training procedures studying the transferability potential of pre-trained features and employing machine learning algorithms specifically designed to be efficient and robust. Also, we address the challenges of limited availability of data resources both in terms of few or not diversified data; here, synthetically generated data, as well as transfer learning and self-supervision are the main methodologies we exploit.

Collaboration with

MaLGA-Imavis joint lab; LCSL @ MaLGa

Unmanned Surface Vehicles (USVs) are autonomous boats that have different applications, ranging from patrolling and monitoring of the waterways for security reasons, to scientific applications such as sampling the water for pollution or biological investigations. One of the major research themes to be tackled, for the widespread adoption of USVs, is the development of reliable obstacle detection and avoidance systems. In this research we investigate data fusion of LIDAR sensors and video-cameras, for efficient detection of potential obstacles, in port areas and open sea.

Collaboration with

UniGe-Graal and ISME