Gait phase detection is essential for clinicians to examine the gait cycles of human test subjects. In this research, a method to detect normal and abnormal gait phases using a single inertial measurement unit (IMU) attached to the shank is proposed. The local maxima, minima and zero crossings of the shank’s angular velocity and acceleration are used to detect heel strike, toe strike and toe off. The proposed method may provide human-understandable insights into the shank’s angular velocity and acceleration to assist clinicians to evaluate the patients’ gaits. The proposed method is compared with an existing method that uses force sensitive resistors (FSRs) placed under the foot. The mean absolute sample difference between the proposed and existing methods is low at around 2 sample difference, equivalent to 20ms difference.
Inertial measurement units (IMUs) have been widely used for many applications such as aircrafts, gaming, animation, and sports. However, the use of IMU in medical diagnosis is still largely unexplored. A great deal of researcher has focused on estimating the pitch angle using accelerometer and gyroscope, while some researchers managed to estimate the roll angle, not many researchers have extended their research to estimate the yaw angle. For those who estimate the yaw angle, the utilization of magnetometer is common, but it requires a more complex algorithm to tackle the magnetic interference. This paper presents an innovative method, called the 2-point error estimation algorithm, to estimate the pitch, roll, and yaw angles using the accelerometer and gyroscope-only. The proposed algorithm is also very computationally efficient to estimate 3-D angles as only two “atan2()” functions are computed throughout the whole walking motion, and seven additions and four multiplications for each angle estimated. The accuracy of the 3-D angles estimated using the IMU is validated against the gold standard Vicon optical motion capture system. The proposed algorithm gives a low average root mean square error of 2.9°, 3.6°, and 4.2° for flexion/extension (pitch), adduction/abduction (roll), and internal/external rotation (yaw) angles of the thigh and shank, respectively. The proposed algorithm can be used to design a low-cost 3-D gait evaluation system.
Commercialized marker-based optical motion capture system like Vicon is very expensive, and not widely available. In the development of inertial measurement units, Vicon is required only for validation so sourcing for expensive system is undesirable. In this paper, a low cost alternative, which is also a marker-based optical motion capture system, is proposed to serve the purpose of validating inertial measurement units. Basically, the proposed system uses a smartphone camera to capture the motions of red markers, then track the coordinates of the markers followed by calculating the angle. The proposed system has a low average root mean square error of 0.83º when the distance between camera and markers is 120cm. The method to synchronize inertial measurement units and camera data is also described in this paper as starting time and sampling rate of the two systems are most likely different.
Indoor positioning systems are drawing ascending attention from the academia as well as industry, motivated by a wide variety of indoor location-based services. Among the existing indoor positioning systems, WiFi-based schemes are more favorable because of the ubiquitous WiFi infrastructures. However, both strong non-line-of-sight scenario and severe ambiguity among location-specific fingerprints prevent most existing WiFi-based indoor positioning systems from achieving centimeter accuracy for localization. In this paper, we propose an indoor positioning system which achieves centimeter accuracy and maintains the performance under non-line-of-sight scenarios using a single pair of off-the-shelf WiFi devices. By harvesting the inherent diversity in WiFi systems, the proposed indoor positioning system formulates a large effective bandwidth to provide a fine-grained location-specific fingerprint with a much higher resolution compared with the fingerprints in most WiFi-based indoor positioning systems. Extensive experiment results in a typical indoor environment show that the centimeter accuracy as well as robustness against environment dynamics can be achieved simultaneously with a large effective bandwidth.
IMU-based stride length estimation remains as a challenging problem despite many researchers have tried to use frequency, vertical acceleration, and/or angle to estimate the stride length. In this paper, we propose 3 new features that can be extracted from a single shank's gyroscope for stride length estimation. We first noticed that the shank's horizontal displacement increases mainly during its extension, while remains almost unchanged during its flexion. The horizontal displacement $s$ during the shank's extension was treated as a linear increment, and therefore a linear kinematic equation $s = 0.5(u + v)t$ could be applied. The initial velocity $u$ is estimated as the average angular velocity during shank's flexion, while the final velocity $v$ is estimated as the summation of angular velocity during the whole stride. The time $t$ is estimated as the duration of the shank's extension rather than the duration of the whole stride. This proposed method had resulted in a low mean absolute error of 4.2cm with a standard deviation of 5.5cm.
In this research, the authors investigate the feasibility of selecting three‐dimensional thigh and shank angles as the features of machine learning methods. Four common machine learning techniques, i.e. random forest, k‐nearest neighbour, support vector machine and perceptron, were compared in terms of accuracy and memory usage so that a real‐time standalone gait diagnosis device can be constructed using low‐end inertial measurement units (IMUs). With proper re‐sampling and normalisation, they discovered that the support vector machine and perceptron resulted in the top two highest accuracies (96–99%) among the four machine learning methods. The memory requirement of the perceptron is the lowest among the machine learning methods. Therefore, perceptron was selected as the classification algorithm for the standalone gait diagnosis device. The trained perceptron was transferred to the thigh and shank's IMUs to process the data locally in real‐time. The constructed standalone gait diagnosis device lit up green or red light emitting diodes when normal or abnormal gaits were detected, respectively. This standalone device was further tested in real‐life and achieved a mean classification accuracy of 96.50%.