Introduction
Preliminaries and Related Work
Clinical Indicators of Abnormal Locomotion Patterns
Martino-Saltzman Model
-
Direct: a simple or uncomplicated trajectory from one location to another one.
-
Pacing: at least three consecutive back-and-forth movements between two locations along very similar paths.
-
Lapping: at least two circular movements between at least three distinct locations.
-
Random: a continuous and aimless movement across numerous locations with multiple directional changes, that generally passes through more than four locations.
Low-Level Motion Indicators
-
Jerk [15] is the rate at which a person’s acceleration changes with respect to time.
-
Sharp angles [6] are defined as vector angles in a trajectory being equal to or more than 90 degrees.
-
Straightness [16] is defined as the ratio of the distance between two consecutive trajectory segments and the distance between the start and end point of these segments.
-
Turning angle [9] is defined as the sum of the absolute angles between any two subsequent lines in a trajectory.
-
Path efficiency [9] is defined as the ratio between the distance from the start to the end of a trajectory and the trajectory length.
IoT Techniques for Detecting Locomotion Anomalies
TraMiner System Overview
Trajectory Segmentation and Visual Feature Extraction
Position Data Cleaning and Trajectory Segmentation
-
We set a threshold \(T_v\) for the maximum possible velocity v of a person moving in the home. If the speed between any two consecutive position records \(\langle r_i, r_{i+1} \rangle \in H\) is higher than \(T_v\), the record \(r_{i+1}\) is considered a noisy reading; hence, it is deleted from H. For the sake of this work, we set \(T_v\) to \(15 \, m/s\).
-
By considering the arrangement of sensors in the test-bed of our experiments, the maximum distance between any two adjacent sensors is below \(3 \, m\). By carefully analyzing the persons trajectories, we observed that some paths spatially deviated from the expected trajectory due to abrupt movements between non-contiguous sensors. In this regard, if the distance between the positions of two consecutive records \(\langle r_i, r_{i+1} \rangle \in H\) exceeds a threshold \(T_d\) (which is set to \(5 \, m\) in this work), we remove \(r_{i+1}\) from H.
Visual Feature Extraction
TRAJ Feature Extraction
SPEED Feature Extraction
DNN Trajectory Classification and Long-Term Analysis
Cloud-Based Model Training
Long-Term Trajectory Analysis
Experimental Evaluation
Dataset
Comparison with State-of-the-Art Methods
State-of-the-Art Numeric Feature Extraction (NFE)
-
Pacing [13] travel pattern, defined in the Martino-Saltzman model: this feature counts the number of observations of this pattern in the last day.
-
Lapping [13] travel pattern, defined in the Martino–Saltzman model: This feature counts the number of observations of this pattern in the last day.
-
Random [13] travel pattern, defined in the Martino-Saltzman model: This feature counts the number of observations of this pattern in the last day.
-
Jerk [15] is computed as the first time derivative of acceleration. This features represents the average jerk observed in the individual’s trajectories in the last day.
-
Straightness [16] represents the average straightness computed on the individual’s trajectories of the last day.
-
Sharp angles [6] feature counts the number of sharp angles observed in the individual’s trajectories during the last day.
State-of-the-Art Visual Feature Extraction (GVFE)
State-of-the-Art DNN (DCNN)
Experimental Setup
Results of NFE Technique
-
The well-known Naive Bayes [49] classifier;
-
Logistic regression classifier [50], relying on a multinomial logistic regression model with a ridge estimator;
-
Multilayer perceptron (MLP) feed-forward artificial neural network algorithm;
-
Support Vector Machines [51]
-
Ripper [53] propositional rule learner;
-
C4.5 [54] decision tree;
-
Random tree [55] classifier.
-
Random forest (RF) [56] classifier.
Class | Measure | Naive Bayes | Logistic regr. | MLP | SVM | |
---|---|---|---|---|---|---|
Cognitively | F\(_1\) score | 0.50 | 0.69 | 0.65 | 0.69 | |
healthy | Precision | 0.56 | 0.55 | 0.53 | 0.53 | |
Recall | 0.45 | 0.91 | 0.84 | 1.00 | ||
MCI | F\(_1\) score | 0.46 | 0.32 | 0.25 | 0.04 | |
Precision | 0.38 | 0.57 | 0.39 | 1.00 | ||
Recall | 0.59 | 0.22 | 0.19 | 0.02 | ||
PwD | F\(_1\) score | 0.00 | n/a | n/a | n/a | |
Precision | 0.00 | n/a | n/a | n/a | ||
Recall | 0.00 | 0.00 | 0.00 | 0.00 | ||
Avg. | F\(_1\) score | 0.32 | n/a | n/a | n/a | |
Precision | 0.31 | n/a | n/a | n/a | ||
Recall | 0.35 | 0.38 | 0.34 | 0.34 | ||
Class | Measure | kNN | Ripper | C4.5 | Rand. tree | RF |
Cognitively | F\(_1\) score | 0.61 | 0.68 | 0.59 | 0.54 | 0.64 |
healthy | Precision | 0.55 | 0.53 | 0.47 | 0.55 | 0.56 |
Recall | 0.68 | 0.93 | 0.79 | 0.54 | 0.74 | |
MCI | F\(_1\) score | 0.43 | 0.21 | 0.00 | 0.34 | 0.39 |
Precision | 0.46 | 0.50 | 0.00 | 0.35 | 0.44 | |
Recall | 0.41 | 0.13 | 0.00 | 0.33 | 0.35 | |
PwD | F\(_1\) score | 0.00 | n/a | 0.00 | 0.00 | 0.08 |
Precision | 0.00 | n/a | 0.00 | 0.00 | 0.20 | |
Recall | 0.00 | 0.00 | 0.00 | 0.00 | 0.05 | |
Avg. | F\(_1\) score | 0.35 | n/a | 0.20 | 0.29 | 0.37 |
Precision | 0.34 | n/a | 0.16 | 0.30 | 0.40 | |
Recall | 0.36 | 0.35 | 0.26 | 0.29 | 0.38 |
Results of Single Trajectory Image Classification
Class | Measure | TRAJ and DCNN | GVFE and DCNN | ||||||
---|---|---|---|---|---|---|---|---|---|
30s | 60s | 120s | 180 | 30s | 60s | 120s | 180s | ||
Cognitively | F\(_1\) score | 0.54 | 0.52 | 0.57 | 0.57 | 0.57 | 0.57 | 0.56 | 0.55 |
healthy | Precision | 0.57 | 0.53 | 0.62 | 0.6 | 0.62 | 0.62 | 0.6 | 0.57 |
Recall | 0.51 | 0.5 | 0.52 | 0.53 | 0.53 | 0.52 | 0.53 | 0.54 | |
MCI | F\(_1\) score | 0.34 | 0.36 | 0.5 | 0.38 | 0.36 | 0.36 | 0.4 | 0.35 |
Precision | 0.33 | 0.36 | 0.4 | 0.37 | 0.34 | 0.34 | 0.39 | 0.35 | |
Recall | 0.35 | 0.37 | 0.41 | 0.39 | 0.38 | 0.38 | 0.41 | 0.36 | |
PwD | F\(_1\) score | 0.12 | 0.13 | 0.1 | 0.12 | 0.11 | 0.056 | 0.13 | 0.11 |
Precision | 0.1 | 0.11 | 0.08 | 0.08 | 0.09 | 0.04 | 0.1 | 0.09 | |
Recall | 0.14 | 0.15 | 0.13 | 0.18 | 0.15 | 0.081 | 0.17 | 0.13 | |
Avg. | F\(_1\) score | 0.33 | 0.34 | 0.39 | 0.36 | 0.35 | 0.33 | 0.36 | 0.34 |
Precision | 0.33 | 0.33 | 0.37 | 0.35 | 0.35 | 0.33 | 0.36 | 0.34 | |
Recall | 0.33 | 0.34 | 0.35 | 0.37 | 0.35 | 0.33 | 0.37 | 0.34 |
Class | Measure | TRAJ and MLP DNN | SPEED and MLP DNN | ||||||
---|---|---|---|---|---|---|---|---|---|
30s | 60s | 120s | 180 | 30s | 60s | 120s | 180s | ||
Cognitively | F\(_1\) score | 0.62 | 0.67 | 0.67 | 0.64 | 0.62 | 0.66 | 0.72 | 0.67 |
healthy | Precision | 0.56 | 0.65 | 0.62 | 0.59 | 0.56 | 0.66 | 0.7 | 0.64 |
Recall | 0.68 | 0.69 | 0.72 | 0.69 | 0.68 | 0.65 | 0.74 | 0.7 | |
MCI | F\(_1\) score | 0.5 | 0.55 | 0.58 | 0.56 | 0.53 | 0.52 | 0.59 | 0.57 |
Precision | 0.46 | 0.5 | 0.52 | 0.52 | 0.51 | 0.46 | 0.55 | 0.56 | |
Recall | 0.56 | 0.62 | 0.65 | 0.6 | 0.56 | 0.61 | 0.64 | 0.59 | |
PwD | F\(_1\) score | 0.35 | 0.34 | 0.47 | 0.33 | 0.37 | 0.42 | 0.39 | 0.35 |
Precision | 0.58 | 0.49 | 0.75 | 0.54 | 0.56 | 0.56 | 0.51 | 0.45 | |
Recall | 0.25 | 0.27 | 0.34 | 0.24 | 0.28 | 0.34 | 0.31 | 0.28 | |
Avg. | F\(_1\) score | 0.49 | 0.52 | 0.57 | 0.51 | 0.50 | 0.53 | 0.56 | 0.53 |
Precision | 0.53 | 0.54 | 0.63 | 0.55 | 0.54 | 0.56 | 0.58 | 0.55 | |
Recall | 0.49 | 0.46 | 0.57 | 0.51 | 0.51 | 0.53 | 0.56 | 0.52 |
Results of Two Input Trajectory Images Classification
Class | Measure | TRAJ+SPEED and MLP DNN | |||
---|---|---|---|---|---|
30s | 60s | 120 | 180s | ||
Cognitively | F\(_1\) score | 0.79 | 0.82 | 0.8 | 0.79 |
healthy | Precision | 0.73 | 0.81 | 0.76 | 0.74 |
Recall | 0.86 | 0.84 | 0.85 | 0.84 | |
MCI | F\(_1\) score | 0.74 | 0.74 | 0.76 | 0.72 |
Precision | 0.67 | 0.64 | 0.68 | 0.69 | |
Recall | 0.84 | 0.86 | 0.86 | 0.78 | |
PwD | F\(_1\) score | 0.49 | 0.56 | 0.58 | 0.52 |
Precision | 0.81 | 0.82 | 0.89 | 0.39 | |
Recall | 0.35 | 0.43 | 0.43 | 0.77 | |
Avg. | F\(_1\) score | 0.67 | 0.71 | 0.71 | 0.67 |
Precision | 0.73 | 0.75 | 0.77 | 0.61 | |
Recall | 0.68 | 0.71 | 0.71 | 0.79 |
Results of Long-Term Trajectory Analysis
Class | Measure | TRAJ+SPEED and MLP DNN | |||
---|---|---|---|---|---|
30s | 60s | 120 | 180s | ||
Cognitively | F\(_1\) score | 0.86 | 0.93 | 0.92 | 0.88 |
healthy | Precision | 0.85 | 0.94 | 0.95 | 0.84 |
Recall | 0.87 | 0.93 | 0.89 | 0.92 | |
MCI | F\(_1\) score | 0.78 | 0.82 | 0.87 | 0.87 |
Precision | 0.69 | 0.74 | 0.8 | 0.83 | |
Recall | 0.9 | 0.93 | 0.93 | 0.9 | |
PwD | F\(_1\) score | 0.57 | 0.67 | 0.83 | 0.61 |
Precision | 0.79 | 0.84 | 0.89 | 0.79 | |
Recall | 0.44 | 0.55 | 0.77 | 0.5 | |
Avg. | F\(_1\) score | 0.73 | 0.81 | 0.87 | 0.78 |
Precision | 0.77 | 0.84 | 0.88 | 0.82 | |
Recall | 0.73 | 0.80 | 0.86 | 0.77 |