featured projects: CovidID Seeing Eye Drone blindBike SCD iSight iVest ULearn
SCD: Senior Collapse Detection
Senior Collapse Detection (SCD) uses the Kinect sensor's fusion of image, audio, and depth sensor data to create a “smart home” system capable of detecting the fall of senior/geriatric individuals in home environments. This research extends previous efforts at fall detection that used body-worn sensors and computer vision techniques. A "fallen" detector that uses the number of joints close to the floor is shown to be the most accurate way of determining if a fall has occurred. Another "fallen" detector is shown that bases fall detections off of the individual's aspect ratios in 3-dimensions and head distance to the floor. A "falling" detector is introduced that uses head velocity, head trajectory, and distance of the head from the floor to detect falls as they happen. The final fall detector discussed looks for falls that result in a user coming to rest on their hands and knees, a special case not caught by the other detectors. We then discuss the actions taken by SCD when a fall is detected, including alerts sent to pre-determined caretakers. A discussion about issues with occlusion, and ways to mitigate its effects are then discussed. We conclude by discussing how we optimized the various thresholds used by the detectors using particle swarm optimization, and how each detector performed across a wide range of fall and non-fall test scenarios.