Tutorial - Train Object Detector

  • This tutorial shows, how to use the Image Analysis Software, Enterprise software to train your own image object detector, in particular for eyes detection. The object detector is trained on the positive (eyes) and negative (randomly choosen background images) training set. The detection process is performed mostly in real-time.

    Please note that this is a tutorial, i.e. a simplified version of a previous real BurgSys's project. The objective is not to achieve the best possible accuracy but to provide easy to understand and easy to configure example. Accuracy of the process can be improved significantly by model optimization and more information you can also get at BurgSys training.

Training process description

  • Loading input data sets. Since the training data set size is quite limited, an extended training set is generated in order to achieve better accuracy.

  • An object detector is trained and stored to disk.

  • Detection process - the stored detector is loaded from the disk and applied to a testing image.1)

  • When the model is trained and validated, the detection can run in real-time, 24 hours a day, 7 days a week and automatically analyse millions of images.1)

    1) Acknowledgment: Image sources provided for free by FreeDigitalPhotos.net. See acknowledgement.