| dc.description.abstract | The research of baby monitor systems has been greatly growing in these years because of newborn infants are usually in hazardous environments are not conscious; On the other hand, young parents concern about the periods of babies growing up. They apply to be learning music, auditory stimulation, etc. to accompany growth of the babies. This thesis aims to develop a system to baby monitoring and musical interaction, system provide a safer living environment, and to be the best playmate for infants. The system starts from the detection of the face information of infant, then establishes limbs ROI model based on the face information. We use dynamic background subtraction algorithm to obtain dynamic information about each limb ROI model, and add some time information to this dynamic information, then the information will be the input of neural network input. Neural network architecture of this thesis is as follows: 8 input neurons, single hidden layer, 10 neurons in the hidden layer and eleven output neurons. The neural network can classify 11 types of state, which includes 6 security status, and five kinds of dangerous status. When the baby is in danger, the system will alarm the guardians immediately. On the other hand, when baby waving limbs, the system will recognize baby’s waving limb, then interact with the baby by changing music via MIDI. System develops with the basement of embedded development board－Cubieboard 4, it has the advantage of small size, low cost and easy to apply in general living environment. In this thesis, the function of danger monitoring detects dangerous events effectively, therefore, guardians can perform other tasks (such as: cooking, hot milk, and so on.) easily; The function of musical interaction can provide the babies can environment to explore for fun, stimulate their hearing ability and physical development.
The performance of the proposed system was verified by nine experimental scenarios, each scenario had specific types of output state, five of which scenario will occur dangerous events, the others are secure scenarios which include musical interaction events. This thesis computes results for each state of confusion matrix, and the results of the indicators we use Precision and Recall, the precision ratio and the recall ratio were 86.6% and 85.7%, respectively. | en_US |