Method for opening and closing screen door using machine learning based on photo pictures of passengers on platform and computing device for the same

• Main Author: KIM HYUN
• Format: Patent
• Language: English; Korean
• Published: 28.06.2021
• Subjects: CALCULATING; CHECKS FOR WINGS; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; DEVICES FOR MOVING WINGS INTO OPEN OR CLOSED POSITION; EQUIPMENT THEREFOR NOT OTHERWISE PROVIDED FOR; FIXED CONSTRUCTIONS; KEYS; LOCKS; PERFORMING OPERATIONS; PHYSICS; RAILWAY SYSTEMS; RAILWAYS; SAFES; TRANSPORTING; WINDOW OR DOOR FITTINGS; WING FITTINGS NOT OTHERWISE PROVIDED FOR, CONCERNED WITH THEFUNCTIONING OF THE WING

A method of training a machine learning device includes: a first acquisition step of selecting, by a computing device, top N passengers with a fastest speed among passengers present in an observation area on a screen obtained by photographing a platform at a reference time point, which is a time point that has elapsed by a predetermined time from a time point at which a screen door installed on the platform is opened, by a camera, and acquiring information on current locations and speeds of the N passengers; a second acquisition step of inputting, by the computing device, input data including the information on the current locations and speeds of the N passengers detected in the observation area to an input layer of the machine learning device, and acquiring a machine suggestion delay, which is a delay time taken from the reference time point to closing of the screen door, from an output layer of the machine learning device; calculating, by the computing device, a difference value obtained by subtracting the acquired machine suggestion delay from an actual delay, which is a time taken from the reference time point to closing of the screen door performed by a first door opening/closing controller that currently operates; and updating, by the computing device, a parameter included in the machine learning device to decrease the difference value. Accordingly, an opening/closing time point of the screen door is controlled by reflecting a density and a movement state of the passengers on the platform. 기계학습기를 학습하는 방법으로서, 컴퓨팅장치가, 카메라가 플랫폼에 설치된 스크린도어가 열린 시각으로부터 소정의 시간만큼 경과한 시각인 기준시각에 상기 플랫폼을 촬영하여 획득한 화면에서 관찰영역에 존재하는 승객들 중 속력이 가장 빠른 상위 N명을 선정하고, 상기 N명의 승객의 현재 위치 및 속도에 관한 정보를 획득하는 제1획득단계, 상기 컴퓨팅장치가, 기계학습기의 입력층에 상기 관찰영역에서 검출된 N명의 승객의 현재 위치 및 속도에 관한 정보를 포함하는 입력데이터를 입력하고, 기계학습기의 출력층으로부터 상기 기준시각으로부터 상기 스크린도어를 닫을 때까지 걸린 딜레이 시간인 기계제안 딜레이를 획득하는 제2획득단계, 상기 컴퓨팅장치가, 상기 기준시각으로부터 현재 운영 중인 제1도어개폐제어기가 상기 스크린도어를 닫을 때까지 걸린 시간인 실제 딜레이로부터 상기 획득한 기계제안 딜레이를 빼서 얻은 차이값을 산출하는 단계, 및 상기 컴퓨팅장치가, 상기 차이값이 감소하도록 상기 기계학습기에 포함된 파라미터를 갱신하는 단계를 포함하는, 기계학습기 학습방법가 공개된다.