Taking Mobile Multi-Object Tracking to the Next Level
Seiten
2014
|
1., Aufl.
Shaker (Verlag)
978-3-8440-2524-8 (ISBN)
Shaker (Verlag)
978-3-8440-2524-8 (ISBN)
- Keine Verlagsinformationen verfügbar
- Artikel merken
Recent years have seen considerable progress in automotive safety and autonomous navigation applications, fueled by the remarkable advance of individual Computer Vision components, such as object detection, tracking, stereo and visual odometry. The goal in such applications is to automatically infer semantic understanding from the environment, observed from a moving vehicle equipped with a camera system. The pedestrian detection and tracking components constitute an actively researched part in scene understanding, important for safe navigation, path planning, and collision avoidance.
Classical tracking-by-detection approaches require a robust object detector that needs to be executed in every frame. However, the detector is typically the most computationally expensive component, especially if more than one object class needs to be detected. A first goal of this thesis was to develop a vision system based on stereo camera input that is able to detect and track multiple pedestrians in real-time. To this end, we propose a hybrid tracking system that combines a computationally cheap low-level tracker with a more complex high-level tracker. The low-level trackers are either based on level-set segmentation or stereo range data together with a point registration algorithm and are employed in order to follow individual pedestrians over time, starting from an initial object detection. In order to cope with drift and to bridge occlusions that cannot be resolved by low-level trackers, the resulting tracklet outputs are fed to a high-level multihypothesis tracker, which performs longer-term data association. With this integration we obtain a real-time tracking framework by reducing object detector applications to fewer frames or even to few small image regions when stereo data is available. Reduction of expensive detector evaluations is especially relevant for the deployment on mobile platforms, where real-time performance is crucial and computational resources are notoriously limited.
To overcome another limitation of a classical tracking-by-detection pipeline, employment only for tracking of objects for which a pre-trained object classifier is available, we propose a tracking-before-detection system that is able to track known and unknown objects robustly, based purely on stereo information. With this approach we track all visible objects in the scene by first segmenting the point cloud into individual objects and associating them to trajectories based on a simple registration algorithm. The core of our approach is a compact 3D representation that allows us to robustly track a large variety of objects, while building up models of their 3D shape online. In addition to improving tracking performance, this representation allows us to detect anomalous shapes, such as carried items on a person’s body. Moreover, classical pedestrian tracking approaches ignore important aspects of human behavior, that should be considered for better scene understanding. Humans are not moving independently, but they closely interact with their surroundings, which includes not only other persons, but also further scene objects. Being able to track not only humans but also their objects, such as child strollers, suitcases, walking aids and bicycles, we propose a probabilistic approach for classifying person-object interactions, which associates objects simultaneously to persons and predicts their interaction type.
In order to demonstrate the capabilities of proposed tracking algorithms, we evaluated them on several challenging video sequences, captured in busy and crowded shopping street environments. As our experiments prove we come closer to the goal of better scene understanding, being able to detect and track multiple objects in the scene in real time and to predict their possible interactions.
Classical tracking-by-detection approaches require a robust object detector that needs to be executed in every frame. However, the detector is typically the most computationally expensive component, especially if more than one object class needs to be detected. A first goal of this thesis was to develop a vision system based on stereo camera input that is able to detect and track multiple pedestrians in real-time. To this end, we propose a hybrid tracking system that combines a computationally cheap low-level tracker with a more complex high-level tracker. The low-level trackers are either based on level-set segmentation or stereo range data together with a point registration algorithm and are employed in order to follow individual pedestrians over time, starting from an initial object detection. In order to cope with drift and to bridge occlusions that cannot be resolved by low-level trackers, the resulting tracklet outputs are fed to a high-level multihypothesis tracker, which performs longer-term data association. With this integration we obtain a real-time tracking framework by reducing object detector applications to fewer frames or even to few small image regions when stereo data is available. Reduction of expensive detector evaluations is especially relevant for the deployment on mobile platforms, where real-time performance is crucial and computational resources are notoriously limited.
To overcome another limitation of a classical tracking-by-detection pipeline, employment only for tracking of objects for which a pre-trained object classifier is available, we propose a tracking-before-detection system that is able to track known and unknown objects robustly, based purely on stereo information. With this approach we track all visible objects in the scene by first segmenting the point cloud into individual objects and associating them to trajectories based on a simple registration algorithm. The core of our approach is a compact 3D representation that allows us to robustly track a large variety of objects, while building up models of their 3D shape online. In addition to improving tracking performance, this representation allows us to detect anomalous shapes, such as carried items on a person’s body. Moreover, classical pedestrian tracking approaches ignore important aspects of human behavior, that should be considered for better scene understanding. Humans are not moving independently, but they closely interact with their surroundings, which includes not only other persons, but also further scene objects. Being able to track not only humans but also their objects, such as child strollers, suitcases, walking aids and bicycles, we propose a probabilistic approach for classifying person-object interactions, which associates objects simultaneously to persons and predicts their interaction type.
In order to demonstrate the capabilities of proposed tracking algorithms, we evaluated them on several challenging video sequences, captured in busy and crowded shopping street environments. As our experiments prove we come closer to the goal of better scene understanding, being able to detect and track multiple objects in the scene in real time and to predict their possible interactions.
| Erscheint lt. Verlag | 30.1.2014 |
|---|---|
| Reihe/Serie | Selected Topics in Computer Vision ; 1 |
| Sprache | englisch |
| Maße | 148 x 210 mm |
| Gewicht | 296 g |
| Einbandart | Paperback |
| Themenwelt | Informatik ► Software Entwicklung ► Mobile- / App-Entwicklung |
| Mathematik / Informatik ► Informatik ► Theorie / Studium | |
| Schlagworte | Close-Range Human Detection • hybrid tracking • Mobile Multi-Object Tracking • Person-Object Interaction • ROI based Object Detection • ROI Extraction • Tracking with Time-Constrained Detection |
| ISBN-10 | 3-8440-2524-3 / 3844025243 |
| ISBN-13 | 978-3-8440-2524-8 / 9783844025248 |
| Zustand | Neuware |
| Haben Sie eine Frage zum Produkt? |
Mehr entdecken
aus dem Bereich
aus dem Bereich
Das große Handbuch zum JavaScript-Framework
Buch | Hardcover (2022)
Rheinwerk (Verlag)
CHF 55,85
Eine umfassende Einführung
Buch | Softcover (2021)
De Gruyter Oldenbourg (Verlag)
CHF 62,90
