In the thessis we have developed a solution for identifying runners from photos taken during sport events. For reaching that goal, we developed a procedure, which first isolates each person on the photo. Then we define a target region (for each person), where we belive that the number on the jersey is located. In the next step we execute further image segmentation in order to isolate single digits. After that we use optical character recognition (TesseractOCR) on each of the isolated digit candidates, to determine if it's a number. In the last steps, we perform further result filtering to remove false positive results. The rest of them is joined into the final number. All the tools we have used and the principles behind them, are presented in the paper. We were focusing on object detectors, since this is the key-point in the whole procedure. We also described the user interface we have created for navigation and result visualisation. In the end we explain problems we were not able to solve and ideas for further development.
|
