Computer Science Graduate Seminar: Vision-based Category-Agnostic Object Tracking for Mobile Robots and Intelligent Vehicles

Thursday, June 27, 2019, 9:30am

Location: UMIC Research Centre, room 025, Mies-van-der-Rohe-Straße 15

Speaker: Aljosa Osep, MSc., i8 (Computer Vision)


The majority of existing vision-based methods perform multi-object tracking in the image domain. Yet, in mobile robotics and autonomous driving scenarios, pixel-precise object localization and trajectory estimation in 3D space are of fundamental importance.

Furthermore, the leading paradigms for vision-based multi-object tracking and trajectory prediction rely heavily on object detectors and effectively limit tracking and motion prediction to a set of predefined classes, while the set of object classes that appear in the real-world is unbounded.

In this talk I will present novel methods for vision-based, category-agnostic multi-object tracking that overcome this limitation. The Category-Agnostic Multi-Object Tracker (CAMOT) leverages recent developments in the area of learning-based object proposal generation and lifts image-based proposal estimates to 3D space in order to estimate trajectories of arbitrary objects.

We further extend CAMOT for the task of video-object proposal generation and demonstrate that by utilizing motion and parallax as consistency filters, can robustly track known and unknown objects and mine large video collections.

I will conclude my talk by demonstrating how such a video-object mining approach can be used for video-object discovery and self-supervised object detector learning.


The computer science lecturers invite interested people to join.