Samuele Gasparrini

A Depth-Based Fall Detection System Using a Kinect® Sensor

We propose an automatic, privacy-preserving, fall detection method for indoor environments, based on the usage of the Microsoft Kinect® depth sensor, in an “on-ceiling” configuration, and on the analysis of depth frames. All the elements captured in the depth scene are recognized by means of an Ad-Hoc segmentation algorithm, which analyzes the raw depth data directly provided by the sensor. The system extracts the elements, and implements a solution to classify all the blobs in the scene. Anthropometric relationships and features are exploited to recognize one or more human subjects among the blobs. Once a person is detected, he is followed by a tracking algorithm between different frames. The use of a reference depth frame, containing the set-up of the scene, allows one to extract a human subject, even when he/she is interacting with other objects, such as chairs or desks. In addition, the problem of blob fusion is taken into account and efficiently solved through an inter-frame processing algorithm. A fall is detected if the depth blob associated to a person is near to the floor. Experimental tests show the effectiveness of the proposed solution, even in complex scenarios.

A Human Activity Recognition System Using Skeleton Data from RGBD Sensors

The aim of Active and Assisted Living is to develop tools to promote the ageing in place of elderly people, and human activity recognition algorithms can help to monitor aged people in home environments. Different types of sensors can be used to address this task and the RGBD sensors, especially the ones used for gaming, are cost-effective and provide much information about the environment. This work aims to propose an activity recognition algorithm exploiting skeleton data extracted by RGBD sensors. The system is based on the extraction of key poses to compose a feature vector, and a multiclass Support Vector Machine to perform classification. Computation and association of key poses are carried out using a clustering algorithm, without the need of a learning algorithm. The proposed approach is evaluated on five publicly available datasets for activity recognition, showing promising results especially when applied for the recognition of AAL related actions. Finally, the current applicability of this solution in AAL scenarios and the future improvements needed are discussed.

Kinect as a Tool for Gait Analysis: Validation of a Real-Time Joint Extraction Algorithm Working in Side View

The Microsoft Kinect sensor has gained attention as a tool for gait analysis for several years. Despite the many advantages the sensor provides, however, the lack of a native capability to extract joints from the side view of a human body still limits the adoption of the device to a number of relevant applications. This paper presents an algorithm to locate and estimate the trajectories of up to six joints extracted from the side depth view of a human body captured by the Kinect device. The algorithm is then applied to extract data that can be exploited to provide an objective score for the “Get Up and Go Test”, which is typically adopted for gait analysis in rehabilitation fields. Starting from the depth-data stream provided by the Microsoft Kinect sensor, the proposed algorithm relies on anthropometric models only, to locate and identify the positions of the joints. Differently from machine learning approaches, this solution avoids complex computations, which usually require significant resources. The reliability of the information about the joint position output by the algorithm is evaluated by comparison to a marker-based system. Tests show that the trajectories extracted by the proposed algorithm adhere to the reference curves better than the ones obtained from the skeleton generated by the native applications provided within the Microsoft Kinect (Microsoft Corporation, Redmond, WA, USA, 2013) and OpenNI (OpenNI organization, Tel Aviv, Israel, 2013) Software Development Kits.

Proposal and Experimental Evaluation of Fall Detection Solution Based on Wearable and Depth Data Fusion

Fall injury issues represent a serious problem for elderly in our society. These people want to live in their home as long as possible and technology can improve their security and independence. In this work we study the joint use of a camera based system and wearable devices, in the so called data fusion approach, to design a fall detection solution. The synchronization issues between the heterogeneous data provided by the devices are properly treated, and three different fall detection algorithms are implemented. Experimental results are also provided, to compare the proposed solutions.

Multimodal Interaction in a Elderly-Friendly Smart Home: A Case Study

This paper discusses different and multimodal user-system interfaces proposed in the framework of a smart home designed to support the independent living of elderly and frail users. It is shown how different technologies and solutions may be complemented and integrated to provide effective interaction both for routine activities of daily living and anomalous situations.

Evaluation and possible improvements of the ANT protocol for home heart monitoring applications

This paper evaluates the use of the wireless ANT protocol as a possible technology to implement a heart monitoring solution for the remote management of patients at their home premises.With respect to common commercial solutions, in which the ANT transceiver transmits single heart rate values, this paper investigates the suitability of the protocol for the transmission of complete ECG traces, which could be a useful feature in the framework of a remote health monitoring solution. Further, a frequency agility capability is also implemented, in order to reduce the impact of RF interferers working in the ISM band, and to improve the ECG transmission performance of the ANT device.

Comparison of RGB-D Mapping Solutions for Application to Food Intake Monitoring

Food intake behaviours are strictly correlated to health, especially for elderly people. Dietary habits monitoring is one of the most challenging activity for researchers in AAL scenario. RGB-D sensors, such as Kinect, provide multiple useful data to perform behavioural analysis in an unobtrusive way. Unfortunately, when using the Kinect sensor, depth and RGB data are not available with the same point of view, and a mapping algorithm is required in order to associate a 3D point to the same pixel in both the RGB and depth frames. In this paper, some techniques for RGB-D mapping of Kinect sensor data are compared, and a proposed implementation is described. Some experimental results in specific conditions are finally provided.

Quality of Kinect Depth Information for Passive Posture Monitoring

The availability of a low cost device incorporating RGB and depth sensors, such as the Microsoft Kinect device, has enabled a plethora of applications and solutions aiming at automatic monitoring of gait and posture, in order to assess the subject’s behavior and possibly prevent future health problems. This paper discusses the quality of the information provided by the depth sensor the Kinect device is equipped with, in order to assess the precision of such information, in different operational contexts, and properly identify the performance that may be reasonably expected from the adoption of such device, in the field of passive monitoring of gait and posture. Some techniques borrowed from the image processing field are also suggested, in order to improve the depth information in specific conditions, by means of a low complexity post-processing of the raw depth data.

An Integrated Approach to Fall Detection and Fall Risk Estimation Based on RGB-Depth and Inertial Sensors

Population ageing is a growing phenomenon, especially in Europe, so researchers are developing Active and Assisted Living solutions to promote ageing in place of elderly people. One of the most critical issues is represented by falls, and the development of fall risk estimation and fall detection tools can increase safety of elderly. The aim of this work is to develop fall risk estimation and fall detection tools using data extracted from wearable and vision-based sensors. First, the synchronization issue between heterogeneous data captured by different sensors is addressed, and a straightforward synchronization procedure based on time delays affecting the samples is provided. Then, fall detection algorithms and a fall risk estimation tool based on Timed Up and Go (TUG) test, exploiting wearable Inertial Measurement Units (IMUs) and an RGB-Depth sensor (Microsoft Kinect) are proposed. The fall detection tool is evaluated on 11 healthy adults simulating 4 different falls and performing 4 activities of daily living, while the TUG is tested on 20 healthy subjects. Encouraging preliminary results are obtained with data acquired in laboratory environment. The tools are privacy preserving since only depth and skeleton information captured by Kinect are processed.

Validation of an optimized algorithm to use Kinect in a non-structured environment for Sit-to-Stand analysis.

The aim of this work is to obtain reliable kinematic measures relative to the execution of the Sit-to-Stand functional evaluation test, by low-cost and widely diffused instrumentation, that even non-experienced users can adopt in non-structured environments, like ambulatory or domestic settings. In particular, the paper refers to a low cost RGB-Depth sensor widely used in the gaming scenario like the Microsoft Kinect sensor. An algorithm is proposed that allows a reliable measure of human motion in a sagittal view. The performance of the proposed algorithm is compared to other two classic commercial algorithms. Results obtained by all the three algorithms have been compared to kinematic results obtained by the use of a stereophotogrammetric system that represents the gold-standard for kinematic measurement of human movement. Average errors of about 4 degrees, both for the trunk/leg angle and for the knee flexion/extension angle, have been obtained by the proposed algorithm and open the way to its possible adoption in non-clinical environments and further applications.