Resource-Constrained Human Presence Detection for Indirect Time-of-Flight Sensors

Face recognition with LiDAR and Time-of-Flight sensors is increasingly used in smartphones to automatically unlock devices when the user is present. The problem is that the presence of the user has to be detected before the energy-intensive face recognition can be started. This work presents a solution by introducing an energy-efficient measurement method for indirect Time-of-Flight sensors, which includes an on-chip processing method to detect the user's presence directly on a Time-of-Flight image sensor. The presented method is based on histogram analysis of direct reflectance images. The unique use of direct reflectance images is a significant enabler as it reduces histogram variations. Therefore our method outperforms conventional histogram detection methods and enables low-power on-chip face detection. It further enables the Time-of-Flight sensor to notify the smartphone application processor to perform the actual face recognition for device unlocking. We show in our evaluation that our method just has a false-positive rate of 17%, while all faces were detected. This is very promising since our measurement method reduces the processing cost which might lead to vast power savings in a future smartphone generation.