Szczegóły publikacji
Opis bibliograficzny
Realizowany komputerowo sposób przetwarzania sekwencji klatek, urządzenie do cyfrowego przetwarzania klatek, produkt w postaci programu komputerowego i system monitorowania — [Computer implemented method for processing a frame sequence, digital frame processing device computer program product and monitoring system] / Akademia Górniczo-Hutnicza im. Stanisława Staszica w Krakowie ; wynalazca: Michał GREGA, Andrzej DZIECH, Krzysztof RUSEK, Andrzej MATIOLAŃSKI, Piotr GUZIK. — Int.Cl.: G06V 20/52(2022.01). — Polska. — tłumaczenie patentu europejskiego ; PL3534294T3 ; Udziel. 2023-09-06 ; Opubl. 2024-01-15. — Zgłosz. nr PL18159762T z dn. 2018-03-02
Autorzy (5)
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 151750 |
|---|---|
| Data dodania do BaDAP | 2024-02-04 |
| Tekst źródłowy | URL |
| Rok publikacji | 2024 |
| Typ publikacji | patent |
| Otwarty dostęp |  |
Streszczenie
A computer implemented method for processing a frame sequence to generate an alarm when a dangerous object is being used including a frame acquisition (11) step, a division into sub-regions (12) step, object identification (13) step and decision making (14) step, in which alert generation (16) step is executed if decision made in decision making step (14) is positive. Division into sub-regions (12) step includes division of the frame into sub-regions that overlap and have minimal dimension of at least 100 pixels. Object identification step (13) is executed in sub-region by sub-region manner and includes processing a frame with a deep neural network trained to detect the dangerous object and returns probability of detection for sub-regions of the frame formed into probability map corresponding to the frame. Decision made in decision making (14) step is based on at least a first criterion and a second criterion applying to the subsequent frames and corresponding probability maps. The first criterion includes probability of detection within a predefined number N of subsequent frames being above predefined value of pmin. The second criterion includes the minimal distance between sub-regions in which probability of detection above predefined value of pmin was observed in subsequent frames being lower than given value dmax.