Research on tplat

Remote Photoplethysmography

Mon, 01 Jan 0001 00:00:00 +0000

Remote Photoplethysmography#

DepthPhys: Near-Infrared Remote Photoplethysmography in Driver Monitoring Systems

Subtle variations in the body can be used to detect underlying physiological signals from camera video data. Existing research in this space is largely focused on extracting physiological signals from traditional two-dimensional planar video data. This project explores how the addition of depth data may be used to augment the results of these approaches. In particular, a novel optical pathway is proposed for camera-based physiological sensing and is validated on a new dataset consisting of seven hours of three-dimensional volumetric near-infrared monochromatic video data. Both two-dimensional and three-dimensional signals are extracted from this dataset and the results of either approach are compared using state-of-the-art camera-based physiological sensing neural methods. These findings provide valuable and novel insight into the utility of three-dimensional signals for camera-based physiological sensing, as well as the capability and limitations of existing camera-based physiological sensing solutions when applied in a driver monitoring system application context.

Subsurface Scattering

Mon, 01 Jan 0001 00:00:00 +0000

Subsurface Scattering#

Harnessing coherent illuminator properties to detect subsurface scattering

The coherent light output by a laser yields unique interference properties in comparison to incoherent light sources. This project harnesses these unique properties to discern bare skin from other categories present on the face within the constraints defined by a driver monitoring software (DMS) application context. More specifically, the project uses various laser speckle imaging techniques on vertical-cavity surface-emitting laser illuminated images to perform clustering of bare skin. Three distinct methods are investigated: beam profile analysis, laser speckle variation analysis and laser speckle contrast imaging (LSCI). All methods are evaluated and a comparison is made highlighting the respective benefits and drawbacks of each technique in a DMS application context. In the final implementation, a convolutional neural network is trained to use temporal LSCI processed images to effectively classify skin. A comparison of models trained on light-emitting diode (LED) versus VCSEL illuminated datasets is presented and a consistent improvement in classification performance is demonstrated for the VCSEL models. Finally, a comprehensive evaluation is provided into the limitations of this model.

IoT Bushfire Detection

Mon, 01 Jan 0001 00:00:00 +0000

IoT Bushfire Detection#

Power Requirements of Ground-Based Bushfire Detection Devices

Recently, there has been increased interested in the use of ground-based Internet of Things (IoT) sensors for bushfire detection applications. Such a system typically comprises sensors and a gateway to collect information and relay it to a control station. A key requirement of such systems is that devices are optimised for the lowest possible power consumption. This paper will focus on the power requirements of such a system and seek to evaluate the feasibility of various power sources. Power consumption is investigated for various detection methods. This is used to calculate lifetime power consumption for various combinations of detection methods (device configurations) under different operating conditions. Finally, research is conducted into several potential power sources. This concludes with a recommendation for a secondary cell with a solar panel for visual detection methods and a primary cell for all other detection methods.