Walking Gait Speed Measurement U sing Privacy Respecting AI Enabled Visual Sensor

The rapidly growing aging population of Canada prefers to age in place despite potentially declining physical and cognitive health. Hence comes the potential for regular health assessments such as gait analysis from the comfort of homes using various smart home health applications. Amongst various available sensors, some have communication delays resulting in inaccurate gait assessment whereas others provide weak data that is only good enough to calculate one or two parameters. Although a surveillance camera is an efficient alternative, it can be considered an intrusion in privacy of the residents increasing their guard against the technology. This paper is the first study of a novel privacy respecting intelligent visual sensor which replaces humans with stick figures in real time video. This modified video provides rich data which can be used for various applications including gait assessment. The methodology proposed successfully calculates the walking speed of the residents with an accuracy of ~86-87
^%
that is limited by the current low and asynchronous frame rate of the sensor. The performance of the sensor is restricted by the currently available processing capacity. The results of this paper confirm the potential of the methodology whilst highlighting some limitations of the device which can be resolved in future technology updates of the sensor.