The Library
Scalable methods for single and multi camera trajectory forecasting
Tools
Styles, Olly (2021) Scalable methods for single and multi camera trajectory forecasting. PhD thesis, University of Warwick.
|
PDF
WRAP_Theses_Styles_2021.pdf - Submitted Version - Requires a PDF viewer. Download (27Mb) | Preview |
Official URL: http://webcat.warwick.ac.uk/record=b3856728
Abstract
Predicting the future trajectory of objects in video is a critical task within computer vision with numerous application domains. For example, reliable anticipation of pedestrian trajectory is imperative for the operation of intelligent vehicles and can significantly enhance the functionality of advanced driver assistance systems. Trajectory forecasting can also enable more accurate tracking of objects in video, particularly if the objects are not always visible, such as during occlusion or entering a blind spot in a non-overlapping multicamera network. However, due to the considerable human labour required to manually annotate data amenable to trajectory forecasting, the scale and variety of existing datasets used to study the problem is limited.
In this thesis, we propose a set of strategies for pedestrian trajectory forecasting. We address the lack of training data by introducing a scalable machine annotation scheme that enables models to be trained using a large Single-Camera Trajectory Forecasting (SCTF) dataset without human annotation. Using newly collected datasets annotated using our proposed methods, we develop two models for SCTF. The first model, Dynamic Trajectory Predictor (DTP), forecasts pedestrian trajectory from on board a moving vehicle up to one second into the future. DTP is trained using both human and machine-annotated data and anticipates dynamic motion that linear models do not capture. Our second model, Spatio-Temporal Encoder-Decoder (STED), predicts full object bounding boxes in addition to trajectory. STED combines visual and temporal features to model both object-motion and ego-motion.
In addition to our SCTF contributions, we also introduce a new task: Multi-Camera Trajectory Forecasting (MCTF), where the future trajectory of an object is predicted in a network of cameras. Prior works consider forecasting trajectories in a single camera view. Our work is the first to consider the challenging scenario of forecasting across multiple non-overlapping camera views. This has wide applicability in tasks such as re-identification and multitarget multi-camera tracking. To facilitate research in this new area, we collect a unique dataset of multi-camera pedestrian trajectories from a network of 15 synchronized cameras. We also develop a semi-automated annotation method to accurately label this large dataset containing 600 hours of video footage. We introduce an MCTF framework that simultaneously uses all estimated relative object locations from several camera viewpoints and predicts the object's future location in all possible camera viewpoints. Our framework follows a Which- When-Where approach that predicts in which camera(s) the objects appear and when and where within the camera views they appear. Experimental results demonstrate the effectiveness of our MCTF model, which outperforms existing SCTF approaches adapted to the MCTF framework.
Item Type: | Thesis (PhD) | ||||
---|---|---|---|---|---|
Subjects: | Q Science > QA Mathematics Q Science > QA Mathematics > QA76 Electronic computers. Computer science. Computer software Q Science > QC Physics T Technology > TK Electrical engineering. Electronics Nuclear engineering T Technology > TL Motor vehicles. Aeronautics. Astronautics |
||||
Library of Congress Subject Headings (LCSH): | Trajectories (Mechanics) -- Data processing, Computer vision, Tracking (Engineering), Pattern recognition systems, Transportation -- Data processing, Automobiles -- Collision avoidance systems | ||||
Official Date: | September 2021 | ||||
Dates: |
|
||||
Institution: | University of Warwick | ||||
Theses Department: | Department of Computer Science | ||||
Thesis Type: | PhD | ||||
Publication Status: | Unpublished | ||||
Supervisor(s)/Advisor: | Sanchez, Victor ; Guha, Tanaya | ||||
Sponsors: | Engineering and Physical Sciences Research Council ; Horizon 2020 (Programme) ; National Research Foundation (Singapore) | ||||
Format of File: | |||||
Extent: | xxiii, 144 leaves : colour illustrations | ||||
Language: | eng |
Request changes or add full text files to a record
Repository staff actions (login required)
View Item |
Downloads
Downloads per month over past year