There exist many methods of using software to detect objects in videos. Before beginning this project, several methods were investigated. All of these methods of object detection and tracking involve some sort of image segmentation – dividing the pixels in the image into meaningful parts. Some of these segmentation algorithms include Mean-Shift, Normalized Cuts, Rayleigh Quotients, and Histogram-based methods (Jepson, 2007). This project focuses on histogram-based methods, which Dalal and Triggs (2005) have shown to be effective and efficient at tracking humans and other rigid objects. Histogram-based object descriptors are used today in cars, security systems, and other systems designed to detect people who may be standing or sitting in a variety of ways. Histogram-based detection is very versatile and is not dependent on the precise shape of an object to function.
Histograms are used to describe target features in an image. The Histogram of Oriented Gradients (HOG) descriptor represents images with a histogram of image gradient orientations and magnitudes. These descriptors are a more accurate representation of the object in an image than simply the colors of the pixels. In order to evaluate the degree to which one histogram is similar to another, the histograms can be compared using a Bhattacharyya coefficient (Derpanis, 2008). This is a crucial element of the project, because it determines the output of the tracking application. Seemann (n.d.) gives a high level description of the implementation of a detection algorithm using the HOG descriptor. Dalal expands upon his original collaboration with Triggs in his PhD thesis, Finding People in Images and Videos (2006).
Dalal, N., & Triggs, B. (2005). Histograms of Oriented Gradients for Human Detection. Retrieved February 21, 2013, from http://lear.inrialpes.fr/people/triggs/pubs/Dalal-cvpr05.pdf
Dalal, N. (2006). Finding People in Images and Videos. Retrieved February 21. 2013, from http://lear.inrialpes.fr/people/dalal/NavneetDalalThesis.pdf
Seemann, E. (n.d.). Computer Vision: Histograms of Oriented Gradients. Retrieved February 21, 2013, from
Derpanis, K. (2008). The Bhattacharyya Measure. Retrieved February 21, 2013, from http://www.cse.yorku.ca/~kosta/CompVis_Notes/bhattacharyya.pdf
Wauthier, F. (2012). Motion Tracking in Image Sequences. Retrieved February 21, 2013 from http://www.cs.berkeley.edu/~flw/tracker/
Jepson, A. (2007). Image Segmentation. Retrieved February 21, 2013 from http://www.cs.toronto.edu/~jepson/csc2503/segmentation.pdf