Project topics and materials
Get Complete Project Material File(s) Now! »
Elderly falls in aging population
Elderly falling down problem is one of the most serious life-threatening events that can occur, especially in the population aged 65 and over. This age group suffers from falls on a yearly basis and half of these elderly do fall repeatedly that also have dramatic psychological, medical and social consequences [6]. Most of these falls are associated with one or more identifiable risk factors (e.g. weakness, unsteady gait, confusion and certain medications), and research has shown that attention to these risk factors can significantly reduce rates of falling. According to the statistics of the Centers for Disease Control and Prevention (CDC), one out of three adults age 65 and older falls each year in the United States, and 61% of these falls occur at homes that cause around 10,000 deaths. However, obtaining a quick assistance after a fall reduces the risk of hospitalization by 26% and the death by 80% [7].
Causes of elderly falls
There are several causes for elderly falls. Some reasons such as age, gender, being unconscious, or suffering from chronic neurological or mental problems, cannot be controlled. Whereas other reasons; such as medications side effects, insufficient vision, poor hearing, or muscle weakness can be controlled or modified. Moreover, the factors associated with falls are classified as personal factors (physiological and neurological functions changes, Alzheimer’s diseases, etc.) and environment factors (loose carpets, wet or slippery floors, poorly constructed steps, etc.). The “Figure 1.5” classifies the most common reasons causing elderly falls according to their origin (environmental or personal) and controllability [8,9]. The presence of more than one cause of falling is common, and several studies have shown that the risk of falling increases dramatically as the number of causes increases.
Fall detection system framework
There have been a lot of fall detection techniques proposed since the early 1990s. It can be said that most fall detection systems runs on the same mechanism. The “Figure 2.2” shows the general framework of a fall detection system [39].
From the research made through the fall detection systems that consist of prototype system, approaches or proposals, it can be categorized into three main categories. The first category of fall detection system involves wearable devices. The next category is fall detection involving ambient analyzer. The last category of fall detection involves motion detection (camera-based). Each system has its own working methodology that has pros and cons regarding the usability, accuracy, privacy, etc. The “Figure 2.3” shows the three categories of fall detection systems.
Wearable sensors
Wearable sensors are electronic devices that are worn by (or implanted in) a person, which are designed to collect biomedical, physiological and activity data [40]. These sensors are typically cheap and small in size making them an attractive solution to low-budget projects. They can benworn on different parts of the body or can be embedded in other devices such as watches, shoes, belts, etc. The main disadvantage of wearable sensors is their high level of obtrusiveness.
Accelerometers are a type of wearable sensors that are widely used in fall detection systems that use threshold-based algorithms to detect fall-related events. A threshold is a limit that when surpassed generates an action in the system (e.g., a fall is detected—caregiver is informed). The most common threshold used with the accelerometer data is the sum vector magnitude of acceleration and the rotation angle information [41]. When a real fall happens, collision between human’s body and ground will produce obvious peak value at the acceleration sum “a” which has magnitude given by the Equation (2.1). = 2√ 2+2+2 (2.1).
Where ax, ay, and az present accelerometer measurements of three axes. The system uses the acceleration sum as the first step to distinguish high intensity movements from others. But normal motions such as jumping or sitting also produce peak values, which mean that additional detection features are required.
The second feature used here is an angle calculated based on acceleration measurements. As human’s motion has low acceleration, it is feasible to get gravity component in each axis by using a low pass filter. If gravity components could be separated before and after human’s fall, then it is possible to calculate the rotation angle of accelerometer coordinate in 3D space, which is also equivalent to the rotation angle of gravity vector relative to fixed coordinate [42]. Coordinate constructed by the accelerometer and the gravity vector is shown in the “Figure 2.4”. The rotation of gravity vector in fixed coordinate is shown in the “Figure 2.5”.
Camera-Based sensing
In the past 10 years, there have been great advances in computer vision and video cameras. It opens up a new branch of methods for fall detection. Indeed, the posture and shape of the faller change drastically in fractions of seconds during a fall [43]. These rapid changes are used to determine if a fall happened. Camera-based systems benefit from these patterns by monitoring the subject’s posture and shape during and after a fall. They use different approaches to detect falls such as monitoring human skeleton. They are robust solutions but their computational cost is unattractive for real time applications. Others systems that are based on simpler features (e.g., the falling angle, vertical projection histogram) are not computationally intensive as the human skeleton based systems but they suffer from a high false alarm rate [44,45].
Privacy issues are one of the biggest worries for camera-based systems, and few solutions to alleviate this problem are known such as sending subjects’ images when an alarm is triggered only. Other possible solutions use deep and infrared cameras [46,47].
Ambient sensing
Ambient sensors are installed in or around the house, appliances and furniture, which are designed to collect different types of data, providing information on the patient’s environment and activities. Pressure sensors are often used because of their non-obtrusiveness and their low cost. The fundamental principle is that the pressure changes regarding the distance between the user and the sensor; the closer the user is to the sensor the higher the pressure is. The low fall detection accuracy (below 90%) is the main disadvantage of these sensors [41].
The rest of this section summarizes the proposed fall detection approaches, the contributions, and the drawbacks/challenges for each research work surveyed in this review.
System evaluation
The results showing the evaluation of a system are usually stored in an (2×2) array known as confusion matrix (see Figure 2.6).
True Positives (TP): number of positive instances that were classified as positive.
True Negatives (TN): number of negative instances that were classified as negative.
False Positives (FP): number of negative instances that were classified as positive.
False Negatives (FN): number of positive instances that were classified as negative.
Table of contents :
Abstract
Acknowledgements
Table of contents
Acronyms
Chapter I – Introduction
I.1 Ambient Intelligence (AmI)
I.2 AmI System Flow
I.3 Aging population
I.4 Elderly falls in aging population
I.5 Causes of elderly falls
I.6 Consequences of elderly falls
I.7 Advances in AmI as a driver for seniors independent living
I.8 Existing AmI projects
I.9 Background of the thesis
I.10 Objectives and structure of the thesis
I.12 Key contributions of this thesis
I.13 Publication list
Chapter II – Fall detection systems and sensing floors: existing prototypes and related works Overview
Section 1: Fall detection systems
II.1.1 Introduction
II.1.2 Fall detection system framework
II.1.2.1 Wearable sensors
II.1.2.2 Camera-Based sensing
II.1.2.3 Ambient sensing
II.1.3 System evaluation
II.1.4 Existing prototypes
II.1.5 Commercially available fall detection systems
II.1.6 Conclusion
Section 2: Sensing floors
II.2.1 Introduction
II.2.2 Existing prototypes
Ph.D. Dissertation High Integrity Personal Tracking Using Fault Tolerant Multi-Sensor Data Fusion
II.2.2.1 Monolithic pressure-sensing floors
II.2.2.2 Modular pressure-sensing floors
II.2.3 Commercially available sensing floors
II.2.4 Conclusion
Section 3: The Inria Nancy – Grand Est sensing floor prototype
II.3.1 Introduction
II.3.2 Origins of the Project
II.3.3 Research projects of the Inria Nancy – Grand Est research center
II.3.4 Intra and inter tiles communications
II.3.5 Inria sensing floor capabilities
II.3.6 Conclusion
General conclusion
Chapter III – Processing and simulating data from the Inria Nancy – Grand Est sensing floor Overview
Section 1: Processing and simulating data from the load sensing floor
III.1.1 Introduction
III.1.2 The load pressure sensors of the Inria Nancy – Grand Est platform
III.1.2.1 Load frequency distribution and sensors calibration
III.1.3 Load data collection
III.1.4 Elder tracking and ADLs recognition using the load data
III.1.4.1 ADLs definition
III.1.4.2 Characteristics of all postures
III.1.4.3 Load data processing
III.1.4.4 Experiments and results
III.1.5 Conclusion
Section 2: Processing and simulating data from the accelerometers
III.2.1 Introduction
III.2.2 The accelerometers of the Inria Nancy – Grand Est platform
III.2.3 Accelerometer data collection
III.2.4 Fall detection using the accelerometer data
III.2.4.1 Accelerometer data preprocessing
III.2.4.2 Features extraction
III.2.4.3 Feature selection
III.2.4.4 Signal change detection
III.2.4.4.1 Results of signal changing detection
III.2.5 Evaluation
III.2.6 Conclusion
Ph.D. Dissertation High Integrity Personal Tracking Using Fault Tolerant Multi-Sensor Data Fusion
Section 3: Merging between the load pressure sensors results and the accelerometers decisions
III.3.1 Introduction
III.3.2 Merging between the load sensors results and the accelerometer decisions
III.3.3 Results
III.3.4 Conclusion
General Conclusion
Chapter IV – Multisensory data fusion with detection and exclusion of faults based on Kullback Leibler divergence
Section 1: State of the art and work positioning
IV.1.1 Multisensory data fusion: definition and objectives
IV.1.2 Data fusion methods
IV.1.2.1 Bayes’ rule
IV.1.2.2 Kalman filter
IV.1.2.3 The information filter for data fusion
IV.1.2.4 Sequential Monte Carlo Methods
IV.1.2.5 Alternatives to probability
IV.1.3 Data fusion consequences
IV.1.4 Fault-tolerant multi-sensors data fusion
IV.1.4.2 Information tools for fault-tolerant data fusion
IV.1.5 Multi-sensors data fusion for positioning systems
IV.1.5.1 PLS related works
Section 2: Proposed approach
IV.2.1 Problem statement
IV.2.2 Inertial measurement unit (IMU)
IV.2.2.1 IMU components, measurements and uses
IV.2.2.2 IMU performance and erroneous measurements
IV.2.3 The load pressure sensors of the Inria Nancy – Grand Est platform
IV.2.4 IMU and load pressure sensors data fusion – Problem formulation
IV.2.4.1 Prediction step using an evolution model
IV.2.4.2 Correction step using an observation model
IV.2.5 Fault detection and exclusion using informational framework
IV.2.5.1 Fault detection
IV.2.5.2 Fault isolation
IV.2.6 Experimentation and validation of results
Conclusion
Ph.D. Dissertation High Integrity Personal Tracking Using Fault Tolerant Multi-Sensor Data Fusion
Chapter V – Conclusion and future works
V.1 Conclusion
V.2 Future works
