Wearable Inertial Sensors for Daily Activity Analysis Based on Adam Optimization and the Maximum Entropy Markov Model

Abstract

Advancements in wearable sensors technologies provide prominent effects in the daily life activities of humans. These wearable sensors are gaining more awareness in healthcare for the elderly to ensure their independent living and to improve their comfort. In this paper, we present a human activity recognition model that acquires signal data from motion node sensors including inertial sensors, i.e., gyroscopes and accelerometers. First, the inertial data is processed via multiple filters such as Savitzky-Golay, median and hampel filters to examine lower/upper cutoff frequency behaviors. Second, it extracts a multifused model for statistical, wavelet and binary features to maximize the occurrence of optimal feature values. Then, adaptive moment estimation (Adam) and AdaDelta are introduced in a feature optimization phase to adopt learning rate patterns. These optimized patterns are further processed by the maximum entropy Markov model (MEMM) for empirical expectation and highest entropy, which measure signal variances for outperformed accuracy results. Our model was experimentally evaluated on University of Southern California Human Activity Dataset (USC-HAD) as a benchmark dataset and on an Intelligent Mediasporting behavior (IMSB), which is a new self-annotated sports dataset. For evaluation, we used the "leave-one-out" cross validation scheme and the results outperformed existing well-known statistical state-of-the-art methods by achieving an improved recognition accuracy of 91.25%, 93.66% and 90.91% when compared with USC-HAD, IMSB, and Mhealth datasets, respectively. The proposed system should be applicable to man-machine interface domains, such as health exercises, robot learning, interactive games and pattern-based surveillance.

Keywords: Adam optimization; accelerometer and gyroscope sensors; inertial sensors; maximum entropy Markov model; multi-fused features.

Figure 1

Flow architecture of proposed human activity recognition model.

Figure 2

Signal Preprocessing. Inertial sensors with unfiltered (unprocessed) and filtered (processed) signals of correct walking activity via (a) Savitzky–Golay, (b) hampel and (c) median filters on the University of Southern California Human Activity Dataset (USC-HAD) dataset.

Figure 3

1D vector plot of statistical features of the walking forward activity log using the USC-HAD dataset.

Figure 4

Accelerometer signal representation via chirp z-transform of the walking forward signal.

Figure 5

Hilbert transform features depicted for x components of the walking forward signal.

Figure 6

Local binary pattern (LBP) applied using signal data. (a) Segment of inertial signal sample, (b) sample values of associate signal, (c) middle value P_c as threshold for associate values P_o, P₁, P₂, …, P₇ and (d) produced LBP code converted into decimal representation.

Figure 7

1-D Walsh–Hadamard transform (WHT) as (a) a WHT signal feature and (b) magnitude of WHT coefficients of the walking forward activity.

Figure 8

First order derivation representation of walking forward and jumping up activities using the USC-HAD dataset.

Figure 9

Adaptive moment estimation (Adam) optimization algorithms with adaptive learning of (a) walking forward and (b) running forward activities using the USC-HAD dataset.

Figure 10

AdaDelta optimization algorithm with adaptive learning of (a) elevator down and (b) elevator up activities using the USC-HAD dataset.

Figure 11

Maximum entropy Markov model algorithm applied to six different activities of the Intelligent Media sporting behavior (IMSB) dataset.

Figure 12

Sensors mounted on human body.

Figure 13

Signals representing wrist motion in badminton (left column) and table tennis (right column).

Figure 14

Signals representing feet movement in football (left column) and basketball (right column).