The University of Southampton
University of Southampton Institutional Repository

Person detection using wide angle overhead cameras

Person detection using wide angle overhead cameras
Person detection using wide angle overhead cameras
In cluttered environments, the overhead view is often preferred because looking down can afford better visibility and coverage. However detecting people in this or any other extreme view can be challenging as there are significant variation in a person's appearances depending only on their position in the picture. The Histogram of Oriented Gradient (HOG) algorithm, a standard algorithm for pedestrian detection, does not perform well here, especially where the image quality is poor. We show that with the SCOVIS dataset, on average, 9 false detections occur per image. We propose a new algorithm where transforming the image patch containing a person to remove positional dependency and then applying the HOG algorithm eliminates 98% of the spurious detections in the noisy images from our industrial assembly line and detects people with a 95% efficiency.

The algorithm is demonstrated as part of a simple but effective person tracking by detection system. This incorporates simple motion detection to highlight regions of the image that might contain people. These are then searched with our algorithm. This has been evaluated on a number of SCOVIS sequences and correctly tracks people approximately 99% of the time. By comparison, the exampled algorithms in the OpenCV are less than approximately 50% efficient.

Finally, we show our algorithm's potential for generalization across different scenes. We show that a classifier trained on the SCOVIS dataset achieves a detection rate of 96% when applied to new overhead data recorded at Southampton. Using the output from this stage to generate labelled 'true positives' data we train a new model which achieves a detection rate of 98%. Both these results compared favourably with the performance of a model trained with manually labelled images. This achieves a detection rate of greater than 99%.
Ahmed, Imran
da04cb7a-e29a-4415-9d4d-5a9d4a2fd55a
Ahmed, Imran
da04cb7a-e29a-4415-9d4d-5a9d4a2fd55a
Carter, John
e05be2f9-991d-4476-bb50-ae91606389da

Ahmed, Imran (2014) Person detection using wide angle overhead cameras. University of Southampton, Physical Sciences and Engineering, Doctoral Thesis, 155pp.

Record type: Thesis (Doctoral)

Abstract

In cluttered environments, the overhead view is often preferred because looking down can afford better visibility and coverage. However detecting people in this or any other extreme view can be challenging as there are significant variation in a person's appearances depending only on their position in the picture. The Histogram of Oriented Gradient (HOG) algorithm, a standard algorithm for pedestrian detection, does not perform well here, especially where the image quality is poor. We show that with the SCOVIS dataset, on average, 9 false detections occur per image. We propose a new algorithm where transforming the image patch containing a person to remove positional dependency and then applying the HOG algorithm eliminates 98% of the spurious detections in the noisy images from our industrial assembly line and detects people with a 95% efficiency.

The algorithm is demonstrated as part of a simple but effective person tracking by detection system. This incorporates simple motion detection to highlight regions of the image that might contain people. These are then searched with our algorithm. This has been evaluated on a number of SCOVIS sequences and correctly tracks people approximately 99% of the time. By comparison, the exampled algorithms in the OpenCV are less than approximately 50% efficient.

Finally, we show our algorithm's potential for generalization across different scenes. We show that a classifier trained on the SCOVIS dataset achieves a detection rate of 96% when applied to new overhead data recorded at Southampton. Using the output from this stage to generate labelled 'true positives' data we train a new model which achieves a detection rate of 98%. Both these results compared favourably with the performance of a model trained with manually labelled images. This achieves a detection rate of greater than 99%.

PDF
Ahmed.pdf - Other
Download (35MB)

More information

Published date: February 2014
Organisations: University of Southampton, Southampton Wireless Group

Identifiers

Local EPrints ID: 363113
URI: http://eprints.soton.ac.uk/id/eprint/363113
PURE UUID: 20ab57ab-bf42-4a3d-98fb-9d0f924e02b7

Catalogue record

Date deposited: 25 Mar 2014 16:32
Last modified: 18 Jul 2017 02:43

Export record

Contributors

Author: Imran Ahmed
Thesis advisor: John Carter

University divisions

Download statistics

Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.

View more statistics

Atom RSS 1.0 RSS 2.0

Contact ePrints Soton: eprints@soton.ac.uk

ePrints Soton supports OAI 2.0 with a base URL of http://eprints.soton.ac.uk/cgi/oai2

This repository has been built using EPrints software, developed at the University of Southampton, but available to everyone to use.

We use cookies to ensure that we give you the best experience on our website. If you continue without changing your settings, we will assume that you are happy to receive cookies on the University of Southampton website.

×