The study, led by Urvashi, Rishu, and Vinay Kukreja from the Centre for Research Impact and Outcome at Chitkara University Institute of Engineering and Technology, follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses, or PRISMA, guidelines to ensure transparency and reproducibility. The researchers began with an initial pool of 411 articles gathered from Google Scholar, IEEE Xplore, ScienceDirect, and Scopus using structured search strings. Through a rigorous multi-stage screening process that included duplicate removal, title and abstract screening, full-text evaluation, and formal quality assessment, the team narrowed the corpus to fifty studies for in-depth analysis. These selected works were then examined for methodological trends, reported recognition accuracies, dataset usage patterns, the pace of research growth, and validation practices.

The scientific premise underlying this field is not new. Psychologists have long observed that emotional states manifest in body movement. Landmark studies showed that people with depression exhibit slower, less expressive walking patterns, and that observers can correctly attribute emotions such as happiness, sadness, anger, and fear to strangers simply by watching them move, even when only point-light displays of the moving joints are visible. What has changed is the technological capability to detect these signals automatically. Advances in sensing technologies, from markerless depth cameras and motion capture systems to wearable inertial measurement units and even floor vibration sensors, combined with the maturation of deep learning, have transformed gait-based emotion recognition from a curiosity of behavioral science into a viable computational discipline.

The review’s quantitative findings reveal a field now dominated by neural approaches. Deep learning methods appeared in forty-seven percent of the analyzed studies, reflecting the field’s decisive shift away from handcrafted features toward representations learned directly from data. Hybrid models, which combine deep learning components with classical machine learning techniques or multiple network architectures, accounted for thirty-one percent of the articles. Purely classical machine learning approaches, once the backbone of the field, now appear in just twenty-two percent of studies. This distribution mirrors broader trends in computer vision and affective computing, where end-to-end learned pipelines have consistently outperformed manually engineered feature descriptors such as gait energy images, joint angle statistics, or discrete cosine transform coefficients.

Perhaps the most striking technical finding concerns which architectures perform best. Graph convolutional neural networks achieved the highest average recognition accuracy at 89.50 percent, followed closely by spatio-temporal graph convolutional networks at 88 percent. The success of these graph-based models is no accident. Human walking is inherently a structured, relational phenomenon: joints are connected by bones, and their trajectories evolve over time. Graph convolutional networks represent the human skeleton as a graph of nodes and edges, allowing the network to learn how local joint configurations and global body posture jointly encode emotion. Spatio-temporal variants extend this idea by adding temporal edges that connect the same joint across consecutive frames, enabling the model to capture both the static shape of the body and the dynamic flow of movement, precisely the two information channels that psychophysics experiments have identified as critical for human perception of emotion from gait.

The review also documents a clear hierarchy among benchmark datasets. The Emotion-Gait dataset dominated the literature, appearing in 24.10 percent of the analyzed studies. The Carnegie Mellon University Motion Capture Database, known as CMU-MoCap, and the Edinburgh Locomotion Mocap Dataset, often paired with a deception detection framework abbreviated as DFD, followed with 12.20 percent each. This concentration highlights both a strength and a vulnerability of the field. Standardized benchmarks enable fair comparison between methods, but heavy reliance on a small number of datasets, many of which contain acted or posed emotional walks rather than naturally occurring emotions, raises questions about how well laboratory performance will translate to real-world deployment. The authors note that validation practices across the field remain uneven, with considerable variation in cross-validation schemes, evaluation metrics, and participant demographics.

The methodological pipeline described across the reviewed studies typically unfolds in several stages. Data is first acquired through one of several modalities: optical motion capture systems providing three-dimensional joint coordinates, consumer depth sensors such as the Microsoft Kinect, wearable accelerometers and gyroscopes mounted on the body or embedded in smartphones and smartwatches, or plantar pressure and floor vibration platforms. Preprocessing then segments individual gait cycles and normalizes for walking speed and body proportions. Feature extraction follows, ranging from kinematic descriptors such as stride length, cadence, joint angles, and movement smoothness, to learned representations produced by convolutional or recurrent layers. Finally, classification maps these features onto emotional categories, most commonly discrete labels drawn from Ekman’s basic emotions or dimensional representations along axes of arousal, valence, and dominance.

Beyond the raw accuracy figures, the review situates gait-based emotion recognition within a wider multimodal ecosystem. Emotion is expressed through faces, voices, physiological signals, and body movement simultaneously, and the most robust systems increasingly fuse information across these channels. The reviewed literature includes hybrid frameworks that combine gait data with electroencephalography, speech, facial expressions, and contextual information. Multimodal fusion offers redundancy and resilience, allowing systems to maintain performance when any single channel is degraded. At the same time, the review points to persistent challenges: small and demographically narrow datasets, difficulty distinguishing felt emotions from merely expressed ones, sensitivity to clothing, viewing angle, and environmental conditions, and the computational demands of real-time processing on mobile and embedded platforms.

The application landscape emerging from the review is strikingly broad. In healthcare, gait emotion analysis shows promise for monitoring depression, anxiety, and mood disorders, offering an unobtrusive complement to subjective clinical assessments; several reviewed studies demonstrated that depression risk could be detected from gait data alone, and that smartwatch signals could support emotion recognition in Parkinson’s disease patients. In human-robot interaction, emotionally aware robots could navigate crowds more safely by anticipating the affective states of pedestrians, adjusting their behavior to the perceived mood of the people around them. Smart homes and assistive technologies could adapt lighting, music, or alerts to a resident’s emotional state inferred from their natural walking. Automotive applications include driver state monitoring through wearable sensors, while security and forensic domains have explored emotional gait cues for investigative purposes.

The review does not shy away from the ethical dimensions of this technology. The authors and the broader literature they survey acknowledge that continuous, passive emotion monitoring raises substantial privacy concerns. Unlike facial recognition, gait can be captured at a distance, without the subject’s awareness or cooperation, and even through clothing-invariant, privacy-preserving pipelines. Several researchers have accordingly explored architectures designed specifically to decouple identity information from emotional content, so that systems can recognize affect without uniquely identifying individuals. Regulatory frameworks for emotional surveillance remain immature, and the review implicitly underscores the need for careful governance as these systems approach commercial deployment, particularly in workplaces, public spaces, and healthcare settings where consent dynamics are asymmetric.

Looking forward, the authors identify several research frontiers. Larger, more diverse, and more ecologically valid datasets are urgently needed, ideally capturing spontaneous rather than performed emotion in real-world environments. Self-supervised and transfer learning approaches could reduce dependence on labeled data, and recent work on learning gait emotion representations from unlabeled skeleton sequences points in this direction. Transformer-based architectures, already displacing convolutional models in many vision tasks, are beginning to appear in gait emotion recognition and may push accuracies beyond the current ceiling near ninety percent. Explainability also demands attention: clinicians and end users will need to understand which kinematic cues drive a system’s judgment of sadness or anger before trusting it in consequential decisions.

For a field that began with psychologists noting that sad people walk differently, the trajectory is remarkable. What this systematic review makes clear is that the machine perception of emotion from movement has crossed a threshold of maturity. With graph-based deep architectures consistently achieving near-human recognition rates, standardized benchmarks in place, and application domains from mental health monitoring to socially aware robotics actively under development, gait-based emotion recognition is poised to move from the laboratory into everyday ambient intelligence. The challenges that remain, in data, validation, and ethics, are substantial, but the review provides the field with something it has lacked: a rigorous, transparent map of where it stands and where it must go next.

The central idea behind the research is deceptively simple: the way a person moves reveals how they feel. Psychologists documented this link as early as the 1980s, showing that observers could reliably identify happiness, anger, sadness and other emotional states from gait alone. Happy walkers tend to bounce, with exaggerated arm swings and an energetic stride, while angry walkers move with heavy, forceful steps and sad walkers drag their feet with slumped posture. Translating this human perceptual ability into an algorithm, however, has proven to be one of the harder problems in affective computing.

The Beijing team’s approach is based on skeleton data, which reduces a walking person to a time-varying set of joint positions, typically dozens of keypoints representing the head, torso, arms and legs. Skeleton-based recognition has several practical advantages over analyzing raw video. It strips away clothing, lighting, background clutter and other confounding factors, protects privacy by discarding facial features and appearance, and compresses each video frame into a compact geometric representation that neural networks can process efficiently. But it also discards information, which means the network must infer emotion purely from the geometry and dynamics of the moving body, often producing skeletal patterns that look nearly identical for visually similar emotional states.

Existing skeleton-based methods have faced three persistent limitations, according to the researchers. The first concerns how the skeleton is modeled spatially. Most graph convolutional networks connect joints according to the physical structure of the body, linking the elbow to the shoulder and the knee to the hip, mirroring anatomical connections. Yet emotions are often expressed through long-distance coordination between body parts that are not physically adjacent. A depressed gait, for example, may combine a dropped head with a shuffling step and reduced arm swing, requiring the model to relate distant joints across the whole body. Physical graphs miss these semantically related connections.

The second limitation is temporal. Many earlier systems rely on local convolutional operations along the time axis, which examine only a small window of consecutive frames at each step. To capture how motion evolves over an entire gait cycle, the network must stack many such layers, and even then its receptive field remains limited. Subtle emotional cues, such as hesitation before a step or the damping of a normally energetic stride, can unfold over long time scales that local windows fail to encompass. The third limitation is the ambiguity problem: emotions such as anxiety and sadness, or excitement and anger, can produce nearly indistinguishable skeleton sequences, and standard classification networks often confuse them.

MSH-GT addresses all three weaknesses through a combination of three modules. The heart of the system is the hierarchical graph-transformer module, which fuses two complementary mechanisms. Hierarchical graph convolutions construct multi-scale semantic connections among joints using what the researchers call a centroid diffusion strategy. Instead of restricting connections to physical neighbors, this strategy treats selected joints as centers and progressively diffuses information outward through the graph, allowing the network to build relationships at multiple semantic scales, from fine local articulations to whole-body coordination patterns. Layered on top of this spatial modeling, Transformer attention mechanisms aggregate information across the entire temporal sequence, giving the model a global view of the gait cycle rather than a succession of narrow local glimpses. Attention weights let the network learn, for each walking sequence, which frames and which joint relationships matter most for distinguishing one emotion from another.

To make the most of different kinds of motion information, the team designed a dual-stream architecture. One stream processes the instantaneous pose of the body, the configuration of joints at each moment, while the other captures motion, the way those joint configurations change over time. A motion-pose spatio-temporal fusion module then combines the two streams, so that static postural cues such as a slumped shoulders and dynamic cues such as stride length and walking speed reinforce one another rather than being analyzed in isolation. This separation echoes the way human observers implicitly combine what a body looks like with how it is moving when judging emotion.

The third innovation targets the ambiguity problem directly. The prototype-based refinement contrastive module learns feature prototypes, essentially representative reference vectors, for each emotional category, and then pulls the features of each training sample toward its correct class prototype while pushing it away from the prototypes of other classes. This contrastive shaping of the feature space increases the margin between confusing emotion pairs, improving the discriminability of features for the hard, borderline samples that most often cause misclassification. In effect, the module teaches the network not just to classify, but to spread similar-looking emotions further apart in its internal representation.

The researchers evaluated MSH-GT on two widely used benchmarks, the Emotion-Gait dataset and the ELMD dataset, under their standard experimental protocols. The model achieved competitive state-of-the-art accuracy on both, outperforming or matching strong graph-based baselines. Crucially, it did so with fewer parameters and fewer floating-point operations, a measure of computational cost, than several competing architectures. That efficiency matters for real-world deployment: gait emotion recognition is envisioned in settings such as robots navigating crowded spaces, where a robot must infer the emotional state of nearby pedestrians quickly enough to adjust its path and behave politely, and in monitoring systems where lightweight models are needed for continuous operation.

The paper situates itself within a rapidly evolving lineage of gait emotion recognition systems. Earlier graph convolutional approaches, including STEP and subsequent adaptive and multiscale graph networks such as STA-GCN, AST-GCN and MSA-GCN, progressively refined how skeletons are represented, while more recent hybrid designs have begun combining graph convolutions with Transformers to exploit both structured spatial modeling and long-range temporal attention. MSH-GT pushes this hybrid strategy further by making the semantic graph itself hierarchical and multi-scale, rather than a fixed physical skeleton, and by adding an explicit mechanism for resolving ambiguous cases.

Practical applications extend well beyond robotics. Researchers have previously shown that multimedia gait analysis can help assess depression in students, and affect-aware systems are being explored for human-vehicle interaction, crowd monitoring and biometric contexts. A privacy-preserving gait model that runs efficiently could support early-warning mental health tools, adaptive smart environments that respond to occupant mood, and surveillance or human-computer interaction systems that read bodily affect without cameras capturing identifying facial images. At the same time, the technology raises familiar ethical questions about emotion surveillance, and the study relies on public datasets collected with informed consent, underscoring the importance of how such data is gathered before any deployment.

The team has released the source code publicly on GitHub, allowing other researchers to reproduce the results, build on the architecture and test it on new data. The work was supported by the National Natural Science Foundation of China. As machines are increasingly asked to share spaces with people, understanding not only where a person is going but how they feel while getting there may become a standard requirement, and models like MSH-GT suggest that the answer may be written in every step we take.