Technical noteA foot-wearable interface for locomotion mode recognition based on discrete contact force distribution
Introduction
Plantar pressure distribution reveals detailed information about foot contact, which is of great value for human gait analysis. Compared with platform based measurement systems (e.g. force plate), wearable contact force sensing systems (e.g. in-shoe sensors) are more portable and convenient for long-term measurement of daily activities, especially in outdoor environments. They allow wider applications with respect to abnormal gait analysis, footwear design, and terrain performance [1]. Various commercial products such as F-Scan® by Tekscan [2], Pedar® by Novel [3] and prototypes of foot wearable measurement systems [1], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13] have been proposed with various applications in sports and gait monitoring. Though wearable plantar pressure measurement systems have already been applied in many fields, there are still potential application spaces in other areas.
One possible application of wearable contact force measurement systems is its integration with robotic assistive devices, e.g. powered lower-limb prostheses. Most current locomotion assistive robots are controlled by finite state machine (FSM) models. Each locomotion mode (e.g. level-walking and stair ascent) has its own control strategy. To better assist humans and achieve natural gait patterns with less energy consumption, these advanced assistive devices should first “know” human movement intentions and then select the appropriate control mode [14], [15], [16]. Therefore, locomotion mode recognition is an important issue for robotic assistive device control. Signals used for locomotion mode recognition can be roughly divided into two main categories. The first category includes bioelectric signals related to human movement, with electromyography (EMG) being the most widely used for locomotion mode recognition [17], [18], [19]. Huang et al. proposed a phase-dependent recognition system to recognize seven locomotion modes using surface EMG signals of lower-limb muscles. The average classification accuracy was 92.2% for able-bodied subjects and 91.6% for amputee subjects. Human body capacitance sensing, which detects leg shape changes caused by muscle contractions during locomotive movements, provides another signal for movement recognition with comparable accuracy [20], [21]. In [21], 10 channels of capacitance signals measured from the lower limb were used to recognize six locomotion tasks. Average recognition accuracies for able-bodied subjects and amputee subjects were 93.6% and 93.4%, respectively. However, for both EMG and capacitance signal measurement, electrodes should directly contact skin, which may cause inconvenience and discomfort. The other kind of sensors used for locomotion mode recognition are on-board mechanical sensors, which includes gyroscopes, accelerometers, goniometers and magnetometers [22], [23], [24]. Varol et al. presented a real-time recognition approach based on prosthesis-implemented sensors. Signals measuring joint angles and angular velocities of the knee and ankle, socket sagittal plane moment, foot forces of heel and ball were collected to realize the recognition of three locomotion modes (standing, sitting, and walking) and transitions between them [22]. However, these signals usually vary with accumulated errors or time-related drifts. To realize satisfactory and reliable recognition performance, classification methods based on multi-sensor fusion were developed in recent years. Huang et al. proposed an EMG-mechanical fusion approach, and obtained much better recognition performance than using only mechanical or EMG signals [25]. Therefore, it is meaningful to add more useful movement information to existing locomotion mode recognition systems in order to further improve the performance. Plantar pressure distribution contains rich information on human gait. However, most existing prosthesis-implemented and orthosis-implemented contact force measurement systems are only used for the detection of gait events or gait phases [26], [27], [28], [29], [30], [31]. To our knowledge, no previous studies have investigated whether the information of contact force can be used to recognize multiple locomotion tasks (e.g. standing, level-ground walking, stair ascent, and stair descent) for robotic assistive device control.
In this paper, we explore the potential of using plantar pressure distribution information for human locomotion mode recognition. To make a systematical analysis, we design a foot-wearable interface, which is comprised of a pair of sensing insoles and transmission circuits. Each of the sensing insoles is integrated with four force sensors to measure discrete contact force distribution signals. To evaluate whether discrete contact force signals can be used for locomotion mode recognition, we propose a classification strategy based on decision tree analysis and linear discriminant analysis, and off-line recognition analysis is performed. To verify the measurement performance of the proposed system, experiments are carried out to investigate system stability of long term working on and adaptability to different ground surfaces. Five able-bodied subjects and one transtibial amputee subject are recruited and asked to perform six types of locomotion tasks, which include sitting, standing, walking, obstacle clearance, stair ascent, and stair descent. Satisfactory recognition performances are obtained for both able-bodied subjects and the amputee subject. These results indicate that discrete contact force distribution signals do provide valuable information for human locomotion mode recognition, and the proposed system can be combined with existing recognition systems based on other sensing information to obtain better recognition performance.
The rest of this paper is organized as follows. Section 2 describes the measurement system in detail. Section 3 presents the locomotion mode recognition strategy. Experiments and results are described in Section 4. The conclusion is made in Section 5.
Section snippets
Discrete contact force distribution measurement system
We presented a foot-wearable interface for locomotion mode recognition based on discrete contact force distribution. The interface was composed of a pair of sensing insoles, signal process circuit modules, and a base station connected to a host computer (Fig. 1(a)). Discrete contact force distribution is measured by sensing insoles placed in users’ shoes. The signal processing module transmits signals to the base station via wireless. The base station rearranges signal sequences measured from
Recognition strategy
Human locomotion activities in daily life are usually sequences of repetitive limb movements [34]. As shown in Fig. 4(a), the measured contact force signals are quasi-periodic, and show different characteristics in different gait phases. Similar results are also observed in EMG signals [19] and capacitance signals [21]. Therefore, a phase-dependent classification method [19], which has been used in some previous studies [19], [21], is appropriate for the recognition of these repetitive
Experimental results
Three experiments were carried out to verify the feasibility of the proposed foot-wearable interface. The first two experiments were designed to investigate system stability of long-term working and adaptability to different ground surfaces. The third experiment was performed to evaluate the recognition performance of six locomotion modes for both able-bodied and amputee subjects.
Conclusion
In this paper, we aim to explore the potential of using signals of discrete contact force distribution for human locomotion mode recognition. The presented measurement system shows stable measurement performance during long term working conditions with no time-related signal drifts. No obvious difference is observed in the measured signals when walking on different ground surfaces. With the phase-dependent recognition strategy, LDA and DTA are used to recognize six locomotion tasks and achieve
Acknowledgments
This work was supported by the National Natural Science Foundation of China (Nos. 61005082 and 61020106005), the Beijing Nova Program (No. Z141101001814001), the Beijing Municipal Science and Technology Project (No. Z151100003715001), the PKU-Biomedical Engineering Joint Seed Grant 2014 and the 985 Project of Peking University (No. 3J0865600). We would like to thank the anonymous reviewers for their valuable suggestions that improved this article.
References (36)
- et al.
An in-shoe device to measure plantar pressure during daily human activity
Med Eng Phys
(2011) - et al.
Repeatability of WalkinSense in shoe pressure measurement system: a preliminary study
Foot
(2012) - et al.
Powered ankle–foot prosthesis to assist level-ground and stair-descent gaits
Neural Netw
(2008) - et al.
Development of EMG-based mode and intent recognition algorithms for a computer-controlled above-knee prosthesis
J Biomed Eng
(1990) - et al.
Adaptive changes of foot pressure in hallux valgus patients
Gait Post
(2012) - et al.
Gait analysis using a shoe-integrated wireless sensor system
IEEE Trans Inf Technol B
(2008) - Tekscan. Tactile pressure measurement, pressure mapping systems, force sensors and measurement systems...
- Novel quality in measurement...
- et al.
A reliable gait phase detection system
Trans Neural Syst Rehab Eng
(2001) - et al.
A wearable biomechatronic interface for controlling robots with voluntary foot movements
IEEE/ASME Trans Mechatron
(2007)
A gait monitoring system based on air pressure sensors embedded in a shoe
IEEE/ASME Trans Mechatron
In-shoe plantar pressure measurement and analysis system based on fabric pressure sensing array
IEEE Trans Inf Technol B
Monitoring of posture allocations and activities by a shoe-based wearable sensor
IEEE Trans Biomed Eng
A flexible sensor technology for the distributed measurement of interaction pressure
Sensors
Upslope walking with a powered knee and ankle prosthesis: initial results with an amputee subject
IEEE Trans Neural Syst Rehab Eng
Cited by (61)
Real-time walking gait terrain classification from foot-mounted Inertial Measurement Unit using Convolutional Long Short-Term Memory neural network
2022, Expert Systems with ApplicationsCitation Excerpt :Although the results are promising, the ability to generalize the classifier to unseen subjects cannot be assessed in this manner. Chen et al. (2015) use insole force sensitive resistors (FSRs) to detect gait modes. They define windows of 200 ms before and after predetermined events, namely foot contact and foot-off, and train an LDA classifier on each window.
A shoe-mounted infrared sensor-based instrumentation for locomotion identification using machine learning methods
2021, Measurement: Journal of the International Measurement ConfederationCitation Excerpt :The past few decades have witnessed tremendous growth in sensing technology and classification algorithms for automated identifications of locomotion terrains. Among various sensing methodology, electromyography (EMG) [8], and mechanical sensors like force myography [9,10], inertial measurement units [11,12], pressure sensors-based insoles [13] are the most popular choices. EMG based classification is highly accurate, but due to the limitations such as – 1) environmental conditions like sweating, humidity, and dust at the location of the electrode distorts the EMG signals [14], 2) EMG signals is limitedly available in amputee with high-level of amputation [11], 3) EMG signal is highly sensitive to the placements of the electrodes [15], 4) High cost of EMG set-up; it is limitedly preferred for terrain classification.
A review of terrain detection systems for applications in locomotion assistance
2020, Robotics and Autonomous SystemsCitation Excerpt :LDA is sometimes further combined with other classifiers. For instance, combined LDA with a decision tree analysis (DTA) classifier are used to identify locomotion mode using only wireless foot insoles [46]. In [47], LDA has been used as a first stage classifier (followed by SVM) for feature extraction of IMU data mounted above the ankle to identify human gait kinematics.
Fabric-Based Insoles With Self-Equipped Sensors for the Foot's Physiological Signal Monitoring
2024, IEEE Sensors JournalDESIGN AND DEVELOPMENT OF A DEVICE FOR REDUCTION OF DATA LOSS AND TIME TAKEN FOR TRANSFERRING THE DATA INTO AN ARTIFICIAL BIOCOMPOSITE SOCKET PROSTHESIS THROUGH ARDUINO
2023, Journal of Mechanics in Medicine and Biology