Northeastern University Background
The difficulties associated with caring for a newborn during their first developmental stages include ensuring safety and passing of developmental milestones set by pediatricians. Baby monitoring systems in the market are bought by parents to directly observe their child and receive alerts when something important happens (crying, waking up, etc.). The visual triggers are often simple (“motion within zone”), which have many false alarms, and fail to catch events of interest like lethal/serious injuries caused by preventable accidents, identification of early signs of neurodevelopmental disorders (eg. Torticollis, cerebral palsy, autism spectrum disorder, sudden infant death syndrome during sleep). While these monitoring devices can provide a level of comfort and security to parents, even the most expensive systems are limited in the information and notification regarding an infant’s activity and developmental process.
There exist very few recent attempts initiated by the computer vision community to automatically perform body pose/posture estimation and movement tracking on videos of infants. These models all heavily rely on having access to both RGB and depth data sequences, which hinders their use in regular webcam-based monitoring systems. On the other hand, recent powerful RGB-based pose estimation models trained on large-scale adult activity datasets have limited success in estimating infant movements due to the significant differences in their body ratios, the complexity of infant poses, and types of their activities. More specifically, publicly available large-scale human pose datasets are predominantly scenes from sports, TV shows, and other daily activities performed by adult humans, and none of these datasets provides exemplars of activities of infants. Additionally, privacy and security considerations hinder the availability of adequate infant images/videos required for training of a robust model with deep structure from scratch. Hence, there is an unmet need for data-efficient and privacy-preserving infant pose and posture recognition models to promote the applications of AI-guided infant motor function screening tools towards early diagnosis and intervention
Researchers at Northeastern have developed an artificial intelligence (AI)-guided cloud-based baby monitoring system, called AiWover. AiWover tracks the baby’s movements in the crib and the playroom, categorizes the poses and postures, and analyzes the baby’s activity and development. AiWover’s advanced AI-based system is based on cutting-edge research and provides accurate monitoring by collecting massive amounts of data on the child’s activities, interpreting the data using advanced AI algorithms, and providing user-friendly summaries and alerts to parents, pediatricians, and developmental specialists. Initial models are focused on infant motion monitoring based on real time body pose assessment, but future components of this ecosystem will include other sensing modalities that parents can integrate in to the secure, privacy-preserving cloud-enabled infant activity monitoring system. Novel private datasets with babies provides AiWover’s algorithms a unique advantage over competitors that rely on adult body/pose imagery. This technology also allows models to learn efficiently from a small number of experimental (real) data supplemented by synthetic data produced by in-house generative models.
Reliable updates through accurate pose and posture tracking allows for smart monitoring of baby’s activity and development
Notifications to alarm care-givers in situations of potential accidents
Identification of early developmental disorders, especially during the first two months after birth
Activity recognition and situational awareness
Privacy-preserving, due to being non-RGB as well its in-situ processing
Unlike existing approaches to in-bed behavior monitoring (i.e. wearable sensors, pressure mats) this system is fully non-contact
Small form factor
Infant monitoring ‑ infant pose/posture monitoring for daily checkup, early motor screening, tele-health/rehabilitation
Sleep posture monitoring for pregnant women
Continuous bed-bound patient monitoring in nursing homes and hospitals
Idea validation through customer discovery
Connecting with prospective investors