Use this URL to cite or link to this record in EThOS:
Title: Global optimisation of multi-camera moving object detection
Author: Yan, Y.
Awarding Body: University of Liverpool
Current Institution: University of Liverpool
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
An important task in intelligent video surveillance is to detect multiple pedestrians. These pedestrians may be occluded by each other in a camera view. To overcome this problem, multiple cameras can be deployed to provide complementary information, and homography mapping has been widely used for the association and fusion of multi‐camera observations. The intersection regions of the foreground projections usually indicate the locations of moving objects. However, many false positives may be generated from the intersections of non‐corresponding foreground regions. In this thesis, an algorithm for multi‐camera pedestrian detection is proposed. The first stage of this work is to propose pedestrian candidate locations on the top view. Two approaches are proposed in this stage. The first approach is a top‐down approach which is based on the probabilistic occupancy map framework. The ground plane is discretized into a grid, and the likelihood of pedestrian presence at each location is estimated by comparing a rectangle, of the average size of the pedestrians standing there, with the foreground silhouettes in all camera views. The second approach is a bottom‐up approach, which is based on the multi‐plane homography mapping. The foreground regions in all camera views are projected and overlaid in the top view according to the multi‐plane homographies and the potential locations of pedestrians are estimated from the intersection regions. In the second stage, where we borrowed the idea from the Quine‐McCluskey (QM) method for logic function minimisation, essential candidates are initially identified, each of which covers at least a significant part of the foreground that is not covered by the other candidates. Then non‐essential candidates are selected to cover the remaining foregrounds by following a repeated process, which alternates between merging redundant candidates and finding emerging essential candidates. Then, an alternative approach to the QM method, the Petrick's method, is used for finding the minimum set of pedestrian candidates to cover all the foreground regions. These two methods are non‐iterative and can greatly increase the computational speed. No similar work has been proposed before. Experiments on benchmark video datasets have demonstrated the good performance of the proposed algorithm in comparison with other state‐of‐the‐art methods for pedestrian detection.
Supervisor: Xu, Ming Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral