Proposal of System to Prevent Sudden Infant Death Syndrome Using Kinect

The sudden, unexplained death of an infant is a tragic family event. Unaccustomed tummy sleeping increases the risk of SIDS. Babies who are used to sleeping on their backs and are placed to sleep on their tummies are 18 times more likely to die from SIDS. In order to prevent SIDS, Babysense and Snuza Hero are commercially available. They are useful products that can be detected at an early stage in the accident baby. However, it may become too late after the accident has occurred. Preventing before the accident occurs in the baby is desirable. In this study, we propose a new system to prevent sudden infant death syndrome using Kinect. This system can detect baby sleeping in a prone position from depth information and image of Kinect. First, this system recognizes the presence of a person from depth. Next, this system recognizes that the person is sleeping in supine or prone position by using the difference in depth with the other. Finally, this system recognizes that a person is sleeping in prone position by face recognition. By preventing baby sleeping in a prone position, the incidence of SIDS is greatly reduced.


Introduction
The sudden, unexplained death of an infant is a tragic family event.Sudden infant death syndrome (SIDS) is just one of several causes of sudden, unexplained death in infancy, but it is the most frequently reported (1)(2)(3) .Unaccustomed tummy sleeping increases the risk of SIDS.Babies who are used to sleeping on their backs and are placed to sleep on their tummies are 18 times more likely to die from SIDS (4)(5) .SIDS is the leading cause of death for infants between 1 month and 12 months of age.SIDS is most common among infants that are 1-4 months old.However, babies can die from SIDS until they are 1 year old.A study was conducted to clarify the preventive strategies for accidents of infants and awareness of SIDS (6) .According to this study, 46 % of the mothers lay their infants on their face.
In order to prevent SIDS, Babysense (7) and Snuza Hero (8) are commercially available.Babysense is a highly sensitive non-touch baby Breathing Movement Monitor.Now parents can get the piece of mind while their baby sleeps.Babysense is intended for detection of respiratory cessation (apnea) in babies.It constantly monitors baby's breathing micro movements through the mattress during sleep and gives an alert to caretakers if breathing stops or becomes irregularly slow, giving critical time to intervene.Snuza Hero is a mobile and easy-to-use movement monitor which clips onto baby's diaper to ensure that normal movement is maintained.Hero detects even the slightest movement and will alert you if your baby's movements are very weak or fall to less than 8 movements per minute.If no movement at all is detected for a period of 15 seconds, Hero will vibrate gently.Often this vibration is enough to rouse the baby, and Hero will revert to monitoring mode.After three vibration/rouse incidents, the Rouse Warning will alert you to the fact that your baby's movements have stopped for 15 seconds on three occasions.If no further movement is detected for another 5 seconds, an alarm will sound to alert you.They are useful products that can be detected at an early stage in the accident baby.However, it may become too late after the accident has occurred.
Preventing before the accident occurs in the baby is desirable.
In this study, we propose a new system to prevent sudden infant death syndrome using Kinect.This system can detect baby sleeping in a prone position from depth information and image of Kinect.By preventing baby sleeping in a prone position, the incidence of SIDS is greatly reduced.

System to Prevent Sudden Infant Death
Syndrome Using Kinect

Structure
The structure of this system is shown in Fig. 1.Kinect is installed on top of the crib.This system gets depth information and image of the baby on the crib using Kinect.The pixel number of depth information and image from Kinect is 640 *480 pixels.
This system estimates the state of the baby from these data as shown in Fig. 2. First, the system recognizes the presence of a person.Next, this system recognizes that a person is sleeping in supine or prone position.Finally, this system recognizes that a person is sleeping in prone position.These methods are described in Section 2.2-2.4.
If this system determines the baby is sleeping in a prone position, this system can prevent SIDS by notifying parents the hazards.

(c) Noise rejection
There is a lot of noise in binary image that was created.Noise rejection is performed in the following procedure.
i. Dilation of 3 pixel ii.Erosion of 5 pixel iii.Dilation of 3 pixel iv.Labeling v.The largest area extraction from areas over 2000 pixel By the above processing, the area of a person is extracted.As an example, the image after noise rejection is shown in Fig. 3d.If the area is not extracted, this system recognizes that there is no person.

Recognition of Sleeping in Supine or Prone Position
The height of sleeping in supine or prone position is lower than that of sleeping sideways or sitting.In this section, this system recognizes that a person is sleeping in supine or prone position by using this difference.Processing similar to Section 2.2 is performed.Threshold of creating a binary image is only changed.In order to detect person except sleeping supine or prone position, threshold is determined to 200 millimeters for baby.In this paper, adults try this system.Threshold is determined to 300 millimeters for adult.By this processing, the area of a person except sleeping supine or prone position is extracted as shown in Fig. 4. If a person is recognized in the processing of Section 2.2 and a person is not recognized in the processing of Section 2.3, this system recognizes that a person sleeping supine or prone position is present.

Recognition of Sleeping in Prone Position
In this section, this system recognizes that a person is sleeping in prone position by face recognition.This recognition uses face recognition, but accuracy of face recognition is not high.By performing face recognition on the image that is extracted around head, accuracy can be improved.Image is obtained from the Kinect.Image of the area of the person obtained in Section 2.2 are extracted.Since baby is about three-head figure, image of 40 % is extracted from the head.In this paper, image of 25 % is extracted from the head because adults try this system.
Result of (g) Fig. 4. Recognition of sleeping in supine or prone position.(a) Supine (b) Result of (a) (c) Prone (d) Result of (c) Fig. 5. Recognition of sleeping in prone position.