Fall risk assessment and early-warning for toddler behaviors at home

Abstract

Accidental falls are the major cause of serious injuries in toddlers, with most of these falls happening at home. Instead of providing immediate fall detection based on short-term observations, this paper proposes an early-warning childcare system to monitor fall-prone behaviors of toddlers at home. Using 3D human skeleton tracking and floor plane detection based on depth images captured by a Kinect system, eight fall-prone behavioral modules of toddlers are developed and organized according to four essential criteria: posture, motion, balance, and altitude. The final fall risk assessment is generated by a multi-modal fusion using either a weighted mean thresholding or a support vector machine (SVM) classification. Optimizations are performed to determine local parameter in each module and global parameters of the multi-modal fusion. Experimental results show that the proposed system can assess fall risks and trigger alarms with an accuracy rate of 92% at a speed of 20 frames per second.

Figure 1.

Flowchart of the proposed fall risk assessment and early-warning system.

Figure 2.

Names and locations of the twenty skeletal joints in a Kinect [12]. Blue line segments represent the set of bones used for body height estimation.

Figure 3.

Posture module (a) push-up climb and (b) pull-up climb.

Figure 4.

Motion module (a) rush-running and (b) high jumping.

Figure 5.

Balance module (a) body sway and (b) body lean.

Figure 6.

Altitude module (a) foot altitude and (b) head altitude.

Figure 7.

Relationship between the distance d and its fall risk p(d) with respect to different values of parameter α.

Figure 8.

Relationship between the rank in the ordered list and its weight with respect to different values of parameter β.

Figure 9.

Least square error optimization to determine the local parameter α in each module. (a) push-up climb; (b) pull-up climb; (c) rush-running; (d) high-jumping; (e) body sway; (f) body lean; (g) foot altitude; and (h) head altitude.

Figure 10.

Global parameter determination based on a recursive grid-based search. (a) diminishing parameter β; and (b) SVM's parameters: C and gamma γ.

Figure 11.

ROC curve of the proposed early-warning system.

Figure 12.

Partial occlusion on the human lower body. (a) before occlusion; (b) entering occlusion: default mode; (c) under occlusion: seated mode; (d) leaving occlusion. And (e)–(h) respective depth images. The average depth values in the dashed rectangle (for the upper body) and in the solid rectangle (for the lower body) were compared.