FIELD: image processing; optical character recognition.
SUBSTANCE: performing video filming, collecting data in three stages: collecting and sorting video material, marking the footage to determine the shooting angle and filtering non-target images, blurred and unsuccessful frames, marking of captured and filtered material for detection of target objects: dog body, dog face, dog nose. Video filming is performed in three stages: nose close-up, muzzle close-up and body filming. During the survey, a register and a questionnaire are kept for unambiguous identification of dogs in a database. Filmed material is transferred to a personal computer and sorted into folders according to a strict hierarchy. Frames are marked for filtration and classification of shooting angles is performed separately for body, muzzle and nose. Frames are marked for solving the task of detection and identification by selecting target objects on the frames: body, muzzle and nose. Obtained video in digital format is stored on a data carrier.
EFFECT: enabling creation of compact and efficient data sets, which make it possible to accelerate training of neural networks, to increase generalizing ability and stability of functioning of neural networks.
1 cl, 11 dwg
