Blink Detection

SDK.detect_blink(self: tfsdk.SDK, tf_image: tfsdk.TFImage, target_face: tfsdk.FaceBoxAndLandmarks) → Tuple[tfsdk.ERRORCODE, tfsdk.BlinkState]

Determine if the eyes of the specified face are open or closed. Note, this function does not work well with eyeglasses, which can be detected using tfsdk.SDK.detect_glasses().

Parameters:

• tf_image - The input tfsdk.TFImage, returned by tfsdk.SDK.preprocess_image().

• target_face - The face on which to run blink detection.

Returns:

• tfsdk.ERRORCODE - The error code. Note, if the face is not frontal facing, an error will be returned.

• tfsdk.BlinkState - The output of the blink prediction.

SDK.detect_blinks(self: tfsdk.SDK, tf_images: List[tfsdk.TFImage], keypoints_vec: List[Annotated[List[tfsdk.Point], FixedSize(106)]]) → Tuple[tfsdk.ERRORCODE, List[tfsdk.BlinkState]]

Determine if the eyes of the specified faces are open or closed. This batch processing method increases throughput when using GPU inference.

Parameters:

• tf_images - A list of input images.

• keypoints_vec - A list of the face landmarks for the faces on which to run blink detection.

Returns:

• tfsdk.ERRORCODE - The error code.

• List[tfsdk.BlinkState] - List of outputs of the blink prediction.

class tfsdk.BlinkState

property is_left_eye_closed

Predicted result for the left eye being closed. Computed as tfsdk.BlinkState.left_eye_score < 0.5.

property is_right_eye_closed

Predicted result for the right eye being closed. Computed as tfsdk.BlinkState.right_eye_aspect_ratio < 0.5.

property left_eye_aspect_ratio

The aspect ratio of the left eye, computed as height / width.

property left_eye_score

Score indicating if the left eye is open or closed. 0 indicates closed, 1 indicates open.

property right_eye_aspect_ratio

The aspect ratio of the right eye, computed as height / width.

property right_eye_score

Score indicating if the right eye is open or closed. 0 indicates closed, 1 indicates open.

to_dict(self: tfsdk.BlinkState) → dict

Return a dictionary representation of the object.

To reduce False Positives, blink detection must be run with several constraints:

• Ensure the face is frontal facing

• The face should be 1 - 3 ft from the camera

• Recommended 640 x 480 resolution or better

• Ensure there is good lighting illuminating the eyes

These constraints are not enforced by the method, and should be enforced by called of SDK for best results.