Able-bodied Users Research Articles

Channel reflection: Knowledge-driven data augmentation for EEG-based brain–computer interfaces

A brain–computer interface (BCI) enables direct communication between the human brain and external devices. Electroencephalography (EEG) based BCIs are currently the most popular for able-bodied users. To increase user-friendliness, usually a small amount of user-specific EEG data are used for calibration, which may not be enough to develop a pure data-driven decoding model. To cope with this typical calibration data shortage challenge in EEG-based BCIs, this paper proposes a parameter-free channel reflection (CR) data augmentation approach that incorporates prior knowledge on the channel distributions of different BCI paradigms in data augmentation. Experiments on eight public EEG datasets across four different BCI paradigms (motor imagery, steady-state visual evoked potential, P300, and seizure classifications) using different decoding algorithms demonstrated that: (1) CR is effective, i.e., it can noticeably improve the classification accuracy; (2) CR is robust, i.e., it consistently outperforms existing data augmentation approaches in the literature; and, (3) CR is flexible, i.e., it can be combined with other data augmentation approaches to further improve the performance. We suggest that data augmentation approaches like CR should be an essential step in EEG-based BCIs. Our code is available online.

Exploring EEG-based motor imagery decoding: a dual approach using spatial features and spectro-spatial Deep Learning model IFNet.

In recent years, the decoding of motor imagery (MI) from electroencephalography (EEG) signals has become a focus of research for brain-machine interfaces (BMIs) and neurorehabilitation. However, EEG signals present challenges due to their non-stationarity and the substantial presence of noise commonly found in recordings, making it difficult to design highly effective decoding algorithms. These algorithms are vital for controlling devices in neurorehabilitation tasks, as they activate the patient's motor cortex and contribute to their recovery. This study proposes a novel approach for decoding MI during pedalling tasks using EEG signals. A widespread approach is based on feature extraction using Common Spatial Patterns (CSP) followed by a linear discriminant analysis (LDA) as a classifier. The first approach covered in this work aims to investigate the efficacy of a task-discriminative feature extraction method based on CSP filter and LDA classifier. Additionally, the second alternative hypothesis explores the potential of a spectro-spatial Convolutional Neural Network (CNN) to further enhance the performance of the first approach. The proposed CNN architecture combines a preprocessing pipeline based on filter banks in the frequency domain with a convolutional neural network for spectro-temporal and spectro-spatial feature extraction. To evaluate the approaches and their advantages and disadvantages, EEG data has been recorded from several able-bodied users while pedalling in a cycle ergometer in order to train motor imagery decoding models. The results show levels of accuracy up to 80% in some cases. The CNN approach shows greater accuracy despite higher instability.

The economic value of augmentative exoskeletons and their assistance

For augmentative exoskeletons that assist able-bodied users, a clear metric of success remains an open question. Here we leverage the Vickrey second-price auction to quantify the economic value added by lower-limb exoskeletons and their assistance. We posited that if exoskeletons provided helpful assistance during a difficult task, this value could be quantified through a lowering of participant auction bids to continue walking. The bidding results were compared across different conditions to determine the economic value of the exoskeleton, bearing in mind also the cost of wearing the added mass of the exoskeleton. Results show that the total value of the exoskeleton and assistance was modest. While most participants found the assistance itself valuable, this value was mostly offset by the extra mass added of wearing the exoskeleton. Our approach provides insight into how exoskeleton wearers may value different aspects of user experience. These results suggest economic value may be a powerful tool in the design and control of exoskeletons that maximize user benefit.

A Wearable Force-Sensitive and Body-Aware Exoprosthesis for a Transhumeral Prosthesis Socket

Upper limb prostheses are commonly mounted to the human residual limb by a passive socket. By this design, the sensitive residual limb is exposed to high reaction wrenches, which can be a source of medical complications. In this article, we introduce an active force-sensitive robotic socket, which carries the prosthesis, offloads the residual limb, and allows guidance via small interaction forces at the same time. We investigate the feasibility of this concept by a force-sensitive and wearable shoulder exoskeleton, called <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">exoprosthesis</i> when being combined with a prosthesis. We provide a first mechatronics prototype, two floating base controllers, and an analysis of the loads acting on the user body. Simulations and experiments confirm the concept and reveal that the wrench at residual limb can be fully compensated for the static case and by <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\approx \text{50\%}$</tex-math></inline-formula> for the investigated motions. Human-in-the-loop tests are successfully performed by three able-bodied users showing the later real-world use case in a complex grasping situation. Overall, we believe that a force-sensitive robotic socket has the potential to advance prosthetics to a new level as it provides an intuitive and seamless user control interface.

Revisiting embodiment for brain–computer interfaces

ABSTRACT Researchers increasingly explore deploying brain–computer interfaces (BCIs) for able-bodied users, with the motivation of accessing mental states more directly than allowed by existing body-mediated interaction. This motivation seems to contradict the long-standing HCI emphasis on embodiment, namely the general claim that the body is crucial for cognition. This paper addresses this apparent contradiction through a review of insights from embodied cognition and interaction. We first critically examine the recent interest in BCIs and identify the extent cognition in the brain is integrated with the wider body as a central concern for research. We then define the implications of an integrated view of cognition for interface design and evaluation. A counterintuitive conclusion we draw is that embodiment per se should not imply a preference for body-mediated interaction over BCIs. It can instead guide research by 1) providing body-grounded explanations for BCI performance, 2) proposing evaluation considerations that are neglected in modular views of cognition, and 3) through the direct transfer of its design insights to BCIs. We finally reflect on HCI’s understanding of embodiment and identify the neural dimension of embodiment as hitherto overlooked.

MiniKers

Automating operations of objects has made life easier and more convenient for billions of people, especially those with limited motor capabilities. On the other hand, even able-bodied users might not always be able to perform manual operations (e.g., both hands are occupied), and manual operations might be undesirable for hygiene purposes (e.g., contactless devices). As a result, automation systems like motion-triggered doors, remote-control window shades, contactless toilet lids have become increasingly popular in private and public environments. Yet, these systems are hampered by complex building wiring or short battery lifetimes, negating their positive benefits for accessibility, energy saving, healthcare, and other domains. In this paper we explore how these types of objects can be powered in perpetuity by the energy generated from a unique energy source - user interactions, specifically, the manual manipulations of objects by users who can afford them when they can afford them. Our assumption is that users' capabilities for object operations are heterogeneous, there are desires for both manual and automatic operations in most environments, and that automatic operations are often not needed as frequently - for example, an automatic door in a public space is often manually opened many times before a need for automatic operation shows up. The energy harvested by those manual operations would be sufficient to power that one automatic operation. We instantiate this idea by upcycling common everyday objects with devices which have various mechanical designs powered by a general-purpose backbone embedded system. We call these devices, MiniKers. We built a custom driver circuit that can enable motor mechanisms to toggle between generating powers (i.e., manual operation) and actuating objects (i.e., automatic operation). We designed a wide variety of mechanical mechanisms to retrofit existing objects and evaluated our system with a 48-hour deployment study, which proves the efficacy of MiniKers as well as shedding light into this people-as-power approach as a feasible solution to address energy needed for smart environment automation.

Nonholonomic Virtual Constraints for Control of Powered Prostheses Across Walking Speeds.

This paper presents a method to design a nonholonomic virtual constraint (NHVC) controller that produces multiple distinct stance-phase trajectories for corresponding walking speeds. NHVCs encode velocity-dependent joint trajectories via momenta conjugate to the unactuated degree(s)-of-freedom of the system. We recently introduced a method for designing NHVCs that allow for stable bipedal robotic walking across variable terrain slopes. This work extends the notion of NHVCs for application to variable-cadence powered prostheses. Using the segmental conjugate momentum for the prosthesis, an optimization problem is used to design a single stance-phase NHVC for three distinct walking speed trajectories (slow, normal, and fast). This stance-phase controller is implemented with a holonomic swing phase controller on a powered knee-ankle prosthesis, and experiments are conducted with an able-bodied user walking in steady and non-steady velocity conditions. The control scheme is capable of representing 1) multiple, task-dependent reference trajectories, and 2) walking gait variance due to both temporal and kinematic changes in user motion.

Ambulation Control System Design for a Hybrid Knee Prosthesis.

Prosthetic knees available to individuals with transfemoral amputation seek to restore functional ability to the user. Passive prosthetic knees are lightweight but can restore only limited, dissipative ambulation activities whereas active knees can provide energy to restore additional ambulation activities such as stair climbing and standing up from a chair. Semi-active prosthetic devices aim to only power a subset of activities and use passive components and control when that power is not necessary. Here, we outline an ambulation control system for a lightweight Hybrid Knee prosthesis that is powered for climbing stairs and passive for other ambulation activities (level-ground walking, walking on an incline, and stair descent). We include preliminary offline and online intent recognition system results for one able-bodied user and one individual with a transfemoral amputation demonstrating low error rates in predicting between active and passive control.

"Mine Works Better": Examining the Influence of Embodiment in Virtual Reality on the Sense of Agency During a Binary Motor Imagery Task With a Brain-Computer Interface.

Motor imagery-based brain-computer interfaces (MI-BCI) have been proposed as a means for stroke rehabilitation, which combined with virtual reality allows for introducing game-based interactions into rehabilitation. However, the control of the MI-BCI may be difficult to obtain and users may face poor performance which frustrates them and potentially affects their motivation to use the technology. Decreases in motivation could be reduced by increasing the users' sense of agency over the system. The aim of this study was to understand whether embodiment (ownership) of a hand depicted in virtual reality can enhance the sense of agency to reduce frustration in an MI-BCI task. Twenty-two healthy participants participated in a within-subject study where their sense of agency was compared in two different embodiment experiences: 1) avatar hand (with body), or 2) abstract blocks. Both representations closed with a similar motion for spatial congruency and popped a balloon as a result. The hand/blocks were controlled through an online MI-BCI. Each condition consisted of 30 trials of MI-activation of the avatar hand/blocks. After each condition a questionnaire probed the participants' sense of agency, ownership, and frustration. Afterwards, a semi-structured interview was performed where the participants elaborated on their ratings. Both conditions supported similar levels of MI-BCI performance. A significant correlation between ownership and agency was observed (r = 0.47, p = 0.001). As intended, the avatar hand yielded much higher ownership than the blocks. When controlling for performance, ownership increased sense of agency. In conclusion, designers of BCI-based rehabilitation applications can draw on anthropomorphic avatars for the visual mapping of the trained limb to improve ownership. While not While not reducing frustration ownership can improve perceived agency given sufficient BCI performance. In future studies the findings should be validated in stroke patients since they may perceive agency and ownership differently than able-bodied users.

Eye gaze controlled adaptive virtual keyboard for users with SSMI

BACKGROUND: Users with Severe Speech and Motor Impairment (SSMI) often use a communication chart through their eye gaze or limited hand movement and care takers interpret their communication intent. There is already significant research conducted to automate this communication through electronic means. Developing electronic user interface and interaction techniques for users with SSMI poses significant challenges as research on their ocular parameters found that such users suffer from Nystagmus and Strabismus limiting number of elements in a computer screen. This paper presents an optimized eye gaze controlled virtual keyboard for English language with an adaptive dwell time feature for users with SSMI. OBJECTIVE: Present an optimized eye gaze controlled English virtual keyboard that follows both static and dynamic adaptation process. The virtual keyboard can automatically adapt to reduce eye gaze movement distance and dwell time for selection and help users with SSMI type better without any intervention of an assistant. METHODS: Before designing the virtual keyboard, we undertook a pilot study to optimize screen region which would be most comfortable for SSMI users to operate. We then proposed an optimized two-level English virtual keyboard layout through Genetic algorithm using static adaptation process; followed by dynamic adaptation process which tracks users’ interaction and reduces dwell time based on a Markov model-based algorithm. Further, we integrated the virtual keyboard for a web-based interactive dashboard that visualizes real-time Covid data. RESULTS: Using our proposed virtual keyboard layout for English language, the average task completion time for users with SSMI was 39.44 seconds in adaptive condition and 29.52 seconds in non-adaptive condition. Overall typing speed was 16.9 lpm (letters per minute) for able-bodied users and 6.6 lpm for users with SSMI without using any word completion or prediction features. A case study with an elderly participant with SSMI found a typing speed of 2.70 wpm (words per minute) and 14.88 lpm (letters per minute) after 6 months of practice. CONCLUSIONS: With the proposed layout for English virtual keyboard, the adaptive system increased typing speed statistically significantly for able bodied users than a non-adaptive version while for 6 users with SSMI, task completion time reduced by 8.8% in adaptive version than nonadaptive one. Additionally, the proposed layout was successfully integrated to a web-based interactive visualization dashboard thereby making it accessible for users with SSMI.

Intuitive real-time control strategy for high-density myoelectric hand prosthesis using deep and transfer learning

Myoelectric hand prostheses offer a way for upper-limb amputees to recover gesture and prehensile abilities to ease rehabilitation and daily life activities. However, studies with prosthesis users found that a lack of intuitiveness and ease-of-use in the human-machine control interface are among the main driving factors in the low user acceptance of these devices. This paper proposes a highly intuitive, responsive and reliable real-time myoelectric hand prosthesis control strategy with an emphasis on the demonstration and report of real-time evaluation metrics. The presented solution leverages surface high-density electromyography (HD-EMG) and a convolutional neural network (CNN) to adapt itself to each unique user and his/her specific voluntary muscle contraction patterns. Furthermore, a transfer learning approach is presented to drastically reduce the training time and allow for easy installation and calibration processes. The CNN-based gesture recognition system was evaluated in real-time with a group of 12 able-bodied users. A real-time test for 6 classes/grip modes resulted in mean and median positive predictive values (PPV) of 93.43% and 100%, respectively. Each gesture state is instantly accessible from any other state, with no mode switching required for increased responsiveness and natural seamless control. The system is able to output a correct prediction within less than 116 ms latency. 100% PPV has been attained in many trials and is realistically achievable consistently with user practice and/or employing a thresholded majority vote inference. Using transfer learning, these results are achievable after a sensor installation, data recording and network training/fine-tuning routine taking less than 10 min to complete, a reduction of 89.4% in the setup time of the traditional, non-transfer learning approach.

Effectiveness of caffeine intake on maximal and sub-maximal physiological markers of exercise intensity among wheelchair users compared to able-bodied individuals

The aim of the current study was to assess the effect of caffeine intake on maximal and sub-maximal physiological markers of exercise intensity and whether group (able-bodied and wheelchair users) moderated these findings. Ten able-bodied men (20.3±2.4 years, length 174.3±5.1 cm, 76.4±9.4 kg) and 9 wheelchair users (29.9±7.1 years, length 164±13 cm, 78.6±20.6 kg) participated in the study. Each participant performed 4 exercise tests. Two ramp exercise tests were used to assess the effect of caffeine intake on maximal values of power output (PO), oxygen uptake (VO2), heart rate (HR), lactate and rate of perceived exertion (RPE): one performed with 6 mg/kg body mass of caffeine ingestion as gelatine capsules, and the second one with placebo. Two constant-load exercise tests at 70% POpeak to volitional exhaustion were used to assess the effect caffeine intake on sub-maximal values of VO2, HR, lactate and RPE: one performed with 6 mg/kg body mass of caffeine ingestion as gelatine capsules, and the second one with placebo. Two way ANOVA revealed that caffeine intake does not affect maximal values of VO2, HR, lactate and RPE (P&gt;0.05). Caffeine intake reduced sub-maximal RPE at 5 min (P&lt;0.05) and 10 min of exercise. Sub-maximal HR at 70% POpeak was higher in caffeine than placebo among wheelchair users (P&lt;0.05). Time to exhaustion at 70% POpeak was significantly longer in caffeine than placebo (P&lt;0.05). VO2max and POpeak were significantly higher among able-bodied than wheelchair users (P&lt;0.05). Caffeine has an effect on sub-maximal RPE and time to volitional exhaustion. Closed-loop exercise mode should be employed in future studies. Greater dosage of caffeine could be used but should not exceed the permitted amount of 12 mg/kg body mass. Wheelchair users should exercise and do more physical activity to enhance VO2max and POpeak.

Biometrics Verification Modality Using Multi-Channel sEMG Wearable Bracelet

Electrical biosignals have the potential for use as biometric authenticators, owing to their ability to facilitate liveness detection and concealed nature. In this work, the viability of using surface electromyogram (sEMG) as a biometric modality for users verification is investigated. A database of multi-channel sEMG signals is created using a wearable armband from able-bodied users. Each user used his/her muscles to form a password that consists of a unique combination of specific hand gestures. A total of 18 features are extracted from the signals in order to distinguish between the users. Several features are extracted in the frequency domain after estimating the power spectral density while using the Welch’s method. Specifically, average frequency, signal power, median frequency, Kurtosis, Deciles, coefficient of dissymmetry, and the peak frequency of the sEMG signal are considered. To further increase the accuracy of the classifier, time domain features are also extracted through segmentation of the signal into 10 segments, and then calculating both the root mean square and length of the signal. Several classifiers that are based on K-nearest Neighbors (KNN), Linear Discernment Analysis (LDA), and Ensemble of Classifiers are constructed, trained, and statistically compared, resulting in an average accuracy in 97.4%, 98.3%, and 98.5%, respectively. False acceptance rate (FAR) and False Rejection Rate (FRR) are estimated for each classifier in order to determine the effectiveness of the biometrics verification system. Although the ensemble classifier accuracy was found to be the highest, the results show that the KNN classifier exhibits a FAR of 0.2% and FRR of 2.9%. Thus, the KNN classifier was found to he the optimum classifier after the extraction of all 18 features. This work demonstrates the usefulness of sEMG as a biometric authenticator in user verification.

Investigating the Accessibility of Voice Assistants With Impaired Users: Mixed Methods Study.

BackgroundVoice assistants allow users to control appliances and functions of a smart home by simply uttering a few words. Such systems hold the potential to significantly help users with motor and cognitive disabilities who currently depend on their caregiver even for basic needs (eg, opening a door). The research on voice assistants is mainly dedicated to able-bodied users, and studies evaluating the accessibility of such systems are still sparse and fail to account for the participants’ actual motor, linguistic, and cognitive abilities.ObjectiveThe aim of this work is to investigate whether cognitive and/or linguistic functions could predict user performance in operating an off-the-shelf voice assistant (Google Home).MethodsA group of users with disabilities (n=16) was invited to a living laboratory and asked to interact with the system. Besides collecting data on their performance and experience with the system, their cognitive and linguistic skills were assessed using standardized inventories. The identification of predictors (cognitive and/or linguistic) capable of accounting for an efficient interaction with the voice assistant was investigated by performing multiple linear regression models. The best model was identified by adopting a selection strategy based on the Akaike information criterion (AIC).ResultsFor users with disabilities, the effectiveness of interacting with a voice assistant is predicted by the Mini-Mental State Examination (MMSE) and the Robertson Dysarthria Profile (specifically, the ability to repeat sentences), as the best model shows (AIC=130.11).ConclusionsUsers with motor, linguistic, and cognitive impairments can effectively interact with voice assistants, given specific levels of residual cognitive and linguistic skills. More specifically, our paper advances practical indicators to predict the level of accessibility of speech-based interactive systems. Finally, accessibility design guidelines are introduced based on the performance results observed in users with disabilities.

The Effect of Context on Small Screen and Wearable Device Users’ Performance - A Systematic Review

Small screen and wearable devices play a key role in most of our daily tasks and activities. However, depending on the context, users can easily experience situationally induced impairments and disabilities (SIIDs). Previous studies have defined SIIDs as a new type of impairment in which an able-bodied user’s behaviour is impaired by the context including the characteristics of a device and the environment. This article systematically reviews the empirical studies on the effect of context on SIIDs. In particular, this review aims to answer the following two research questions: Which contextual factors have been examined in the literature that can cause SIIDs and how different contextual factors affect small screen and wearable device users’ performance. This article systematically reviews 187 publications under a framework that has five factors for context analysis: physical, temporal, social, task, and technical contexts. This review shows that a significant amount of empirical studies have been conducted focusing on some factors such as mobility but there still are some factors such as social factors that need to be further considered for SIIDs. Finally, some factors have shown to have significant impact on users’ performance such as multitasking but not all factors has been empirically demonstrated to have an effect on users’ performance.

Using Generic Upper-Body Movement Strategies in a Free Walking Setting to Detect Gait Initiation Intention in a Lower-Limb Exoskeleton

In recent years, lower-limb exoskeletons have been marketed to become a possible alternative to wheelchairs for people with walking impairments or paralysis. However, most assistive exoskeletons rely on constraining control strategies based on remote controls or torso tilting events. One approach to build more intuitive control interfaces would be to exploit knowledge on human motor control, and the coordination between upper and lower limb movements during gait events, such as the anticipatory postural adjustments that precede gait initiation. In this study, it was hypothesized that generic trunk and arm movements preceding gait in a free walking setting in able-bodied users can still be retrieved while wearing an assistive lower-limb exoskeleton that constrains their leg movements. This was evaluated on a group of eight unimpaired participants, and analysis of upper-body wearable IMU signals showed similarity of postural adjustments between the free and exoskeleton-constrained settings. Additionally, a classification architecture showed that the walking state of the robot can be correctly triggered based on free-walking data gathered from all participants with an accuracy rate of 95%. This suggests that interlimb coordination still exists in a constrained setting, and could pave the road towards the elaboration of more natural controls for assistive lower-limb exoskeletons.

Mechanotactile Sensory Feedback Improves Embodiment of a Prosthetic Hand During Active Use.

There have been several advancements in the field of myoelectric prostheses to improve dexterity and restore hand grasp patterns for persons with upper limb loss, including robust control strategies, novel sensory feedback, and multifunction prosthetic terminal devices. Although these advancements have shown to improve prosthesis performance, a key element that may further improve acceptance is often overlooked. Embodiment, which encompasses the feeling of owning, controlling and locating the device without the need to constantly look at it, has been shown to be affected by sensory feedback. However, the specific aspects of embodiment that are influenced are not clearly understood, particularly when a prosthesis is actively controlled. In this work, we used a sensorized simulated prosthesis in able-bodied participants to investigate the contribution of sensory feedback, active motor control, and the combination of both to the components of embodiment; using a common methodology in the literature, namely the rubber hand illusion (RHI). Our results indicate that (1) the sensorized simulated prosthesis may be embodied by able-bodied users in a similar fashion as prosthetic devices embodied by persons with upper limb amputation, and (2) mechanotactile sensory feedback might not only be useful for improving certain aspects of embodiment, i.e., ownership and location, but also may have a modulating effect on other aspects, namely sense of agency, when provided asynchronously during active motor control tasks. This work may allow us to further investigate and manipulate factors contributing to the complex phenomenon of embodiment in relation to active motor control of a device, enabling future study of more precise quantitative measures of embodiment that do not rely as much on subjective perception.

Persona Cards: Introducing Users With Diverse Needs To Enhance Accessibility Teaching

Location: Dewberry Hall Most analysts and designers who follow a formal degree in Information Technology or Computer Science are not familiar with users with diverse needs. Therefore, they tend to implement applications targeted at a regular, able-bodied user. These applications are often inaccessible for users with disabilities. To improve the students' awareness about accessibility needs, enhancing their knowledge acquisition, training and build capacity, students should become more familiarized with diverse needs users may have. Learning about universal design, accessibility considerations and guidelines that render interactive systems more accessible is key in a world where interactive technologies are widespread.

The dynamic grouping keyboard: a general keyboard optimization approach for users with motor impairment

An important feature of software keyboards is its ability to flexibly adapt the interface. In this paper, we present the Dynamic Grouping Keyboard, as a general approach to optimize text entry for users with motor impairments. Each time after a key is selected, the Dynamic Grouping Keyboard adapts the grouping of keys to optimize the interaction cost, such as the number of keystrokes and the distance of cursor movement, while maintaining the key locations to preserve keyboard layout. We identify the technical challenge, and propose a two-step algorithm for optimization: The first step estimates the amount of information that the current step should provide, while the second step minimizes the interaction cost by applying dynamic programming. Two lab studies with twelve able-bodied users were conducted, in which three keyboard cases for motor impairment were tested. Results showed that compared to an optimal static grouping (ambiguous) keyboard, the Dynamic Grouping Keyboard reduced interaction cost by 23–75%, and improved text entry rate by 16–71% across all tested cases. Subjective user feedback showed consistent preference for the Dynamic Grouping Keyboard due to its faster input speed and less fatigue.

Obstacle Recognition using Computer Vision and Convolutional Neural Networks for Powered Prosthetic Leg Applications.

In this work we combine computer vision and a machine learning algorithm, Convolutional Neural Networks (CNNs), to identify obstacles that powered prosthetic leg users might encounter during walking. Our motivation is that powered prosthetic legs could react in synchronicity with their users by recognizing and anticipating the terrain in front of them. We focus on identifying stairs and doors that are within the visual field of a person. To achieve this, we used a compact CNN architecture to optimize image processing for real-time applications. We built and tested a wearable system prototype that included a camera mounted on a pair of glasses and a single-board computer. The prototype was used by able-bodied users to collect and label obstacle and non-obstacle videos, which were used later to train the CNN. In validation, the system was able to recognize around 90% of obstacles across different indoor and outdoor scenarios. The accuracy achieved and the practicality of the prototype shows the potential of computer vision and machine learning in the field of powered prosthetic legs.