https://github.com/gheinrich/bibnumber

Recognize bib numbers from racing photos
https://github.com/gheinrich/bibnumber

Last synced: 6 days ago
JSON representation

Recognize bib numbers from racing photos

• Host: GitHub
• URL: https://github.com/gheinrich/bibnumber
• Owner: gheinrich
• Created: 2015-07-23T12:11:14.000Z (over 9 years ago)
• Default Branch: master
• Last Pushed: 2018-04-03T14:03:39.000Z (over 6 years ago)
• Last Synced: 2024-08-03T04:05:46.176Z (4 months ago)
• Language: C++
• Size: 9.94 MB
• Stars: 117
• Watchers: 20
• Forks: 39
• Open Issues: 15
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• awesome-ocr - Recognize bib numbers from racing photos

README

        
# Bibnumber: automatic bib number recognition from racing photos

Bibnumber automatically recognizes bib number from racing photos. As an example, consider this photo:

![Example](samples/0017-IMG_0035.JPG)

Calling bibnumber on this example will produce the following output:

     [ 38 46 54 69 164 773 775]

Bibnumber also creates small images out of the recognized bibs:

![Bib1](samples/0017-IMG_0035.out/bib-00000-0164.png)

![Bib2](samples/0017-IMG_0035.out/bib-00001-0775.png)

![Bib3](samples/0017-IMG_0035.out/bib-00002-0773.png)

![Bib4](samples/0017-IMG_0035.out/bib-00003-0069.png)

![Bib5](samples/0017-IMG_0035.out/bib-00004-0038.png)

![Bib6](samples/0017-IMG_0035.out/bib-00005-0046.png)

![Bib7](samples/0017-IMG_0035.out/bib-00006-0054.png)

## Implementation overview

Bibnumber is tuned for high accuracy rather than high recall. Therefore it is not unusual for a bib to be missing in the output however wrong bib numbers are unusual. Since the algorithm precision is low on single-digit bib numbers, those are ignored.

The general pipeline is as follows:

* Canny edge detection

* Stroke Width Transform (to locate text)

* Text candidates filtering (to detect possible bib numbers)

* Rotation (to correct moderate skew in bib images)

* OCR (to convert text image to bib number)

The precision of the algorithm is low for short bibnumbers (less than 3 digits). Therefore, the processing can split into three stages:

* in the first stage, only bib numbers with 3 digits or more are detected

* in the second stage, the 3+-digit bib images are used to train a HOG+SVM bib detector

* in the third stage, the HOG+SVM detector is used to detect 2-digit bib numbers

## Installation

The following libraries should be installed to build Bibnumber:

* Tesseract

* OpenCV 2.4.x

* Boost

* Leptonica

To build the project:

	make -C bibnumber/Debug

## Command line

	./bibnumber [-train dir] [-model svmModel.xml] image_file|folder_path|csv_ground_truth_file

    

Bibnumber can either process whole directories or individual images files. To automatically quantify the quality of bib detections, a ground truth .csv file can be used and Bibnumber will display the F-score when done.

In order to train a HOG+SVM bib detector from a number of bib images, the training directory may be specified and Bibnumber will create the SVM model.xml file, which can then be used in a second pass to detect shorter bib numbers (2 letters) with better accuracy.