Detail publikace

Single template object detector based on histogram of oriented gradients

Originální název

Single template object detector based on histogram of oriented gradients

Anglický název

Single template object detector based on histogram of oriented gradients

Jazyk

en

Originální abstrakt

Most of the current image object detection algorithms use very large data sets for their training and these methods are also optimized for those big data sets. Unfortunately, in many cases it is very costly or even impossible to collect large data sets for training (e.g. in medicine, astronomy, and other fields). In this paper a new approach based on Dalal's Histogram of Oriented Gradients (HOG) [3] is introduced. It is devoted for training from a single training template and is optimized to achieve reasonable accuracy with this limited training set. The accuracy is validated on 100 images, where half of them contains positive and the other half negative images. The accuracy achieved is 98%.

Anglický abstrakt

Most of the current image object detection algorithms use very large data sets for their training and these methods are also optimized for those big data sets. Unfortunately, in many cases it is very costly or even impossible to collect large data sets for training (e.g. in medicine, astronomy, and other fields). In this paper a new approach based on Dalal's Histogram of Oriented Gradients (HOG) [3] is introduced. It is devoted for training from a single training template and is optimized to achieve reasonable accuracy with this limited training set. The accuracy is validated on 100 images, where half of them contains positive and the other half negative images. The accuracy achieved is 98%.

BibTex


@inproceedings{BUT117997,
  author="Pavel {Novák} and Radim {Burget} and Jan {Karásek} and Malay Kishore {Dutta}",
  title="Single template object detector based on histogram of oriented gradients",
  annote="Most of the current image object detection algorithms use very large data sets for their training and these methods are also optimized for those big data sets. Unfortunately, in many cases it is very costly or even impossible to collect large data sets for training (e.g. in medicine, astronomy, and other fields). In this paper a new approach based on Dalal's Histogram of Oriented Gradients (HOG) [3] is introduced. It is devoted for training from a single training template and is optimized to achieve reasonable accuracy with this limited training set. The accuracy is validated on 100 images, where half of them contains positive and the other half negative images. The accuracy achieved is 98%.",
  booktitle="International Conference on Telecommunications and Signal Processing (TSP)",
  chapter="117997",
  doi="10.1109/TSP.2015.7296367",
  edition="38",
  howpublished="electronic, physical medium",
  year="2015",
  month="july",
  pages="508--512",
  type="conference paper"
}