Histogram of oriented gradients

from Wikipedia, the free encyclopedia
Left the original and right the HOG calculated and visualized with Scikit-image

The histogram of oriented gradients (HOG) is a tool for extracting features in image processing . It is used in particular to detect people in photos. In a work at Udacity in 2017, the use of detecting moving vehicles was also demonstrated in a video. A training process is always required for object recognition. The size of the training data set has an influence on the quality of the object recognition. At HOG, this training requires significantly fewer examples and can be performed much faster than e.g. B. be carried out with Haar wavelets .

The basic idea is that the appearance and shape of objects within a photo can be represented by the distribution of the local intensity or the arrangement of the edges even without detailed knowledge of the positions of edges or corners. The image is therefore broken down into sub-areas and the orientations of all edges are determined for each sub-area and their number is saved as a histogram.

Another application is that in this way the similarity between different images can be determined - similar to the scale-invariant feature transform (SIFT).

Robert K. McConnell described the basic process in a patent as early as 1986, the concept and the name "Histogram of oriented gradients" became known through a publication by Navneet Dalal and Bill Triggs in 2005, which was at the Institut national de recherche at that time Research in informatique et en automatique .

software

Visualization created with Dlib of a trained HOG detector for face recognition.

In free software - libraries for image processing as Scikit-image , OpenCV and Dlib includes methods for calculating the HOG.

literature

  • Dalal, Navneet; Triggs, Bill. Histograms of oriented gradients for human detection. In: Computer Vision and Pattern Recognition, 2005. CVPR 2005. IEEE Computer Society Conference on. IEEE, 2005. pp. 886-893. doi : 10.1109 / CVPR.2005.177

Individual evidence

  1. ↑ a b Navneet Dalal and Bill Triggs: Histograms of Oriented Gradients for Human Detection. Retrieved January 5, 2019 .
  2. ↑ Mithi: Vehicle Detection with HOG and Linear SVM. In: Towards Data Science. March 28, 2017, accessed January 5, 2019 .
  3. ^ Mithi Sevilla: A vehicle detection and tracking pipeline with OpenCV, histogram of oriented gradients (HOG), and support vector machines (SVM). In: GitHub . December 25, 2018, accessed January 5, 2019 .
  4. ↑ Davis King: Dlib 18.6 released: Make your own object detector! February 3, 2014, accessed January 7, 2019 .
  5. ^ Robert K. McConnell: Method of and apparatus for pattern recognition . In: Patent US4567610A . July 22, 1982 (English, google.com [accessed January 5, 2019]).
  6. ^ Histogram of Oriented Gradients. In: scikit-image - docs. Retrieved January 5, 2019 .
  7. ↑ HOGDescriptor Struct Reference. In: OpenCV. Retrieved January 5, 2019 .
  8. ^ Histogram of Oriented Gradients. In: Learn OpenCV. Retrieved January 5, 2019 .
  9. ^ Image processing. In: dlib C ++ Library. Retrieved January 5, 2019 .