얼굴 검출 도구를 미세 조정하려면 다음 단계를 따르세요.

1. 얼굴 회전 각도 분석을 구성하세요.
   
   1. 얼굴 회전 각도를 확인할 필요가 있다면 Analyze face rotation angle 매개변수를 예로 설정하세요 (1). 이 설정은 특정 얼굴 검색 시 지정된 값보다 회전 및 기울기 각도가 큰 결과를 필터링할 수 있게 합니다 (유사한 얼굴 검색 참조).
      
   2. Face rotation pitch (얼굴 회전 피치) (°)필드에서 얼굴의 상하 기울기 각도를 설정하세요 (3). 값은 [0; 90] 범위 내여야 합니다.
   3. Face rotation roll (얼굴 회전 롤) (°) 필드에서 얼굴의 좌우 기울기 각도를 설정하세요 (4). 값은 [0; 90] 범위 내여야 합니다.
   4. Face rotation yaw from (얼굴 회전 요우 시작) (°) 필드에서 얼굴이 오른쪽이나 왼쪽으로 회전할 수 있는 최소 각도를 설정하세요 (5). 값은 [-90; 90] 범위 내여야 합니다.

   5. Face rotation yaw to (얼굴 회전 요우 끝) (°) 필드에서 얼굴이 오른쪽이나 왼쪽으로 회전할 수 있는 최대 각도를 설정하세요 (6). 값은 [-90; 90] 범위 내여야 합니다.

2. Select a face detection algorithm (2):
   
   1. ALG1—high speed, low accuracy.
   2. ALG2—average speed, average accuracy.
   3. ALG3—low speed, high accuracy.
3. Set up false mask detections filtering:
   1. To apply filters, set Yes for the False mask detections filtering parameter (7).
   2. Set the minimal percentage of probability which makes the additional algorithm identify a track as a masked face in the Minimum filtering threshold for face mask detection field (13). If the algorithm takes a decision that the track relates to a masked face with a probability value lower than the specified threshold, the track will be ignored. Set the value by trial-and-error, values over 30 are recommended.
4. Set the minimum threshold value for mask detection (9). Set a value by trial-and-error, values over 70 are recommended.
5. Set the minimum quality of a face image for recognition with a mask (10, 마스크 검출 구성 참조). Set the value by trial-and-error, values over 30 are recommended.
6. Set the minimum quality of a face image for recognition without a mask (11). Set the value by trial-and-error, values over 50 are recommended.
7. If it is necessary to filter out false positives, set the minimum percentage of probability which makes the algorithm identify a track as a human face in the Minimum filtering threshold field (12, 얼굴 검출 구성 참조). If the algorithm takes a decision that the track relates to a face with a probability value lower than the specified threshold, the track will be ignored. Set the value by trial-and-error, values over 50 are recommended.
8. If it is necessary for the detection tool to use a color frame for processing, then set Yes for the Process color frames parameter (15). By default, a black and white frame is processed.
9. Configure the repeated face recognition ignoring:
   1. If it is necessary to ignore repeated recognition of the same face, then set Yes for the Ignore repeated recognitions parameter (8).
   2. In the Repeated recognitions similarity threshold field (16), set the similarity threshold of a face with the previous recognized ones in percentage from 0 to 100. If the similarity threshold is below the specified value, then the face will be recognized as a new one.
   3. In the Period of ignoring repeated recognitions field (14), set the period in minutes during which new recognized faces will be compared with the previous ones to identify similarities. The value should be in the range [0; 30].
10. Click the Apply button.

Fine-tuning the Face detection tool is now complete.