Hand Actions Research Articles

Activation of Cerebellum, Basal Ganglia and Thalamus During Observation and Execution of Mouth, hand, and foot Actions

Humans and monkey studies showed that specific sectors of cerebellum and basal ganglia activate not only during execution but also during observation of hand actions. However, it is unknown whether, and how, these structures are engaged during the observation of actions performed by effectors different from the hand. To address this issue, in the present fMRI study, healthy human participants were required to execute or to observe grasping acts performed with different effectors, namely mouth, hand, and foot. As control, participants executed and observed simple movements performed with the same effectors. The results show that: (1) execution of goal-directed actions elicited somatotopically organized activations not only in the cerebral cortex but also in the cerebellum, basal ganglia, and thalamus; (2) action observation evoked cortical, cerebellar and subcortical activations, lacking a clear somatotopic organization; (3) in the territories displaying shared activations between execution and observation, a rough somatotopy could be revealed in both cortical, cerebellar and subcortical structures. The present study confirms previous findings that action observation, beyond the cerebral cortex, also activates specific sectors of cerebellum and subcortical structures and it shows, for the first time, that these latter are engaged not only during hand actions observation but also during the observation of mouth and foot actions. We suggest that each of the activated structures processes specific aspects of the observed action, such as performing internal simulation (cerebellum) or recruiting/inhibiting the overt execution of the observed action (basal ganglia and sensory-motor thalamus).

Deciphering Functional Connectivity Differences Between Motor Imagery and Execution of Target-Oriented Grasping.

This study aimed to delineate overlapping and distinctive functional connectivity in visual motor imagery, kinesthetic motor imagery, and motor execution of target-oriented grasping action of the right hand. Functional magnetic resonance imaging data were obtained from 18 right-handed healthy individuals during each condition. Seed-based connectivity and multi-voxel pattern analyses were employed after selecting seed regions with the left primary motor cortex and supplementary motor area. There was equivalent seed-based connectivity during the three conditions in the bilateral frontoparietal and temporal areas. When the seed region was the left primary motor cortex, increased connectivity was observed in the left cuneus and superior frontal area during visual and kinesthetic motor imageries, respectively, compared with that during motor execution. Multi-voxel pattern analyses revealed that each condition was differentiated by spatially distributed connectivity patterns of the left primary motor cortex within the right cerebellum VI, cerebellum crus II, and left lingual area. When the seed region was the left supplementary motor area, the connectivity patterns within the right putamen, thalamus, cerebellar areas IV-V, and left superior parietal lobule were significantly classified above chance level across the three conditions. The present findings improve our understanding of the spatial representation of functional connectivity and its specific patterns among motor imagery and motor execution. The strength and fine-grained connectivity patterns of the brain areas can discriminate between motor imagery and motor execution.

Simultaneous assessment and training of an upper-limb amputee using incremental machine-learning-based myocontrol: a single-case experimental design

BackgroundMachine-learning-based myocontrol of prosthetic devices suffers from a high rate of abandonment due to dissatisfaction with the training procedure and with the reliability of day-to-day control. Incremental myocontrol is a promising approach as it allows on-demand updating of the system, thus enforcing continuous interaction with the user. Nevertheless, a long-term study assessing the efficacy of incremental myocontrol is still missing, partially due to the lack of an adequate tool to do so. In this work we close this gap and report about a person with upper-limb absence who learned to control a dexterous hand prosthesis using incremental myocontrol through a novel functional assessment protocol called SATMC (Simultaneous Assessment and Training of Myoelectric Control).MethodsThe participant was fitted with a custom-made prosthetic setup with a controller based on Ridge Regression with Random Fourier Features (RR-RFF), a non-linear, incremental machine learning method, used to build and progressively update the myocontrol system. During a 13-month user study, the participant performed increasingly complex daily-living tasks, requiring fine bimanual coordination and manipulation with a multi-fingered hand prosthesis, in a realistic laboratory setup. The SATMC was used both to compose the tasks and continually assess the participant’s progress. Patient satisfaction was measured using Visual Analog Scales.ResultsOver the course of the study, the participant progressively improved his performance both objectively, e.g., the time required to complete each task became shorter, and subjectively, meaning that his satisfaction improved. The SATMC actively supported the improvement of the participant by progressively increasing the difficulty of the tasks in a structured way. In combination with the incremental RR-RFF allowing for small adjustments when required, the participant was capable of reliably using four actions of the prosthetic hand to perform all required tasks at the end of the study.ConclusionsIncremental myocontrol enabled an upper-limb amputee to reliably control a dexterous hand prosthesis while providing a subjectively satisfactory experience. The SATMC can be an effective tool to this aim.

Using a Gesture Recognition Modeling approach to improve graphical user interface

The gesture recognition technology based on visual detection employs image processing for detection, segmentation, tracking, and recognition of hand gestures. Static and dynamic gesture recognition are the two types of gesture recognition. In this paper, we present our hand gesture interface consisting of three main layers: Hand Gesture Interface, Mapping Gesture, Action, and Graphical User Interface with our approach named IGMA (Improving Graphical User Interface Based on Gesture Recognition Modeling Approach). IGMA organized in two stages: the modeling stage and the code generation stage. The model is Domain Specific Model; it allows action and gesture modeling. At the code generation stage, a specific code generator transforms all models to final source code using a framework specifically designed for a given target platform. Finally, to simplify the implementation of this method, we propose defining a process for improving user interface control based on gesture recognition. We validate our approach by performing experiments on a gastronomy application. The result is an adaptation framework that can guide software engineers in developing Graphical User Interface Based on Gesture Recognition.

O hand, where art thou? Mapping hand location across the visual field during common activities

Humans employ visually-guided actions during a myriad of daily activities. These ubiquitous but precise manual actions rely on synergistic work between eye and hand movements. During this close cooperation between hands and eyes, the hands persist in sight in a way which is unevenly distributed across our visual field. One common assertion is that most hand actions occur in the lower visual field (LVF) because the arms are anatomically lower than the head, and objects typically rest on waist-high table surfaces. While experimental work has shown that humans are more efficient at reaching for and grasping targets located below their visual midline (Goodale and Danckert, Exp Brain Res 137:303–308, 2001), there is almost no empirical data detailing where the hands lie in the visual fields during natural hand actions. To build a comprehensive picture of hand location during natural visually guided manual actions, we analyzed data from a large-scale open-access dataset containing 100 h of non-scripted manual object interactions during domestic kitchen tasks filmed from a head-mounted camera. We found a clear vertical visual asymmetry with hands located in the lower visual scene (LVS) in more than 70% of image frames, particularly in ipsilateral space. These findings provide the first direct evidence for the established assumption that hands spend more time in the lower than in the upper visual field (UVF). Further work is required to determine whether this LVF asymmetry differs across the lifespan, in different professions, and in clinical populations.

Sound-action symbolism in relation to precision manipulation and whole-hand grasp usage.

It has been suggested that actions can provide a fruitful conceptual context for sound symbolism phenomena, and that tight interaction between manual and articulatory processes might cause that hand actions, in particular, are sound-symbolically associated with specific speech sounds. Experiment 1 investigated whether novel words, built from speech sounds that have been previously linked to precision or power grasp responses, are implicitly associated with perceived actions that present precision manipulation or whole-hand grasp tool-use or the corresponding utilisation pantomimes. In the two-alternative forced-choice task, the participants were more likely to match novel words to tool-use actions and corresponding pantomimes that were sound-symbolically congruent with the words. Experiment 2 showed that the same or even larger sound-action symbolism effect can be observed when the pantomimes present unfamiliar utilisation actions. Based on this we propose that the sound-action symbolism might originate from the same sensorimotor mechanisms that process the meaning of iconic gestural signs. The study presents a novel sound-action phenomenon and supports the view that hand-mouth interaction might manifest itself by associating specific speech sounds with grasp-related utilisations.

YOLO Series for Human Hand Action Detection and Classification from Egocentric Videos

Hand detection and classification is a very important pre-processing step in building applications based on three-dimensional (3D) hand pose estimation and hand activity recognition. To automatically limit the hand data area on egocentric vision (EV) datasets, especially to see the development and performance of the "You Only Live Once" (YOLO) network over the past seven years, we propose a study comparing the efficiency of hand detection and classification based on the YOLO-family networks. This study is based on the following problems: (1) systematizing all architectures, advantages, and disadvantages of YOLO-family networks from version (v)1 to v7; (2) preparing ground-truth data for pre-trained models and evaluation models of hand detection and classification on EV datasets (FPHAB, HOI4D, RehabHand); (3) fine-tuning the hand detection and classification model based on the YOLO-family networks, hand detection, and classification evaluation on the EV datasets. Hand detection and classification results on the YOLOv7 network and its variations were the best across all three datasets. The results of the YOLOv7-w6 network are as follows: FPHAB is P = 97% with TheshIOU = 0.5; HOI4D is P = 95% with TheshIOU = 0.5; RehabHand is larger than 95% with TheshIOU = 0.5; the processing speed of YOLOv7-w6 is 60 fps with a resolution of 1280 × 1280 pixels and that of YOLOv7 is 133 fps with a resolution of 640 × 640 pixels.

Underactuated embedded constraints gripper for grasping in toxic environments

In this paper a soft gripper is proposed and designed to achieve some of the 17 Sustainable Development Goals (SDG) described by United Nations (UN) and in particular SDG3, SDG8, SDG 9 and SDG 12. In fact, the presented gripper is conceived for application in the waste industry for helping or partially replacing human operations which could lead to risks or hazards for human health. The device can artificially reproduce the action of human hands allowing a more sustainable work, focusing the attention on worker’s health. Also the design characteristics are oriented to sustainability by using eco-friendly materials. Furthermore, the device is an underactuated soft gripper with modular elements and without sensors. There are no electronic components, and the damageable and non-recyclable parts are minimized. After the description of gripper and mechanical analysis, three different configurations (wearable, with extension and mounted on a cobot) are presented where it is possible to notice that the ends of the gripper (the fingers) are far from the most delicate and less recyclable components such as the motor. Thus, thanks to the modularity of the fingers, it is easy to replace damaged fingers: they have a lower environmental impact than electronic components. In this way, the presented project falls in “the circular design for sustainability” in robotics.

Image-recognition-based system for precise hand function evaluation

Hand dysfunction caused by hand injuries, strokes, or other neurological degenerative diseases such as cervical spondylosis is being increasingly reported. Currently, hand function assessments for diagnosis or rehabilitation are primarily based on qualitative scales, which are subjective and may vary considerably depending on the expertise of the attending clinician. Although wearable sensors and computer vision techniques have been proposed to obtain quantitative hand movement information, both have limitations. In this study, a multiview video tracking and recording system was set up using high-speed cameras and mapping of actual hand movements. The state-of-the-art software DeepLabCut was used to obtain precise 2D and 3D finger joint positions. Kinematic parameters, such as movement count, period, phase, and Pearson coefficient were used to characterize hand movement based on the relative distance-time curves of finger joints. Experimental results in a clinical setting showed that this video-based image-recognition neural network method can accurately distinguish healthy from dysfunctional hand movements. The proposed system is inexpensive, easy to set up and use, and exhibits high accuracy. Thus, it can revolutionize medical hand motion analysis and spur the development of automated quantitative systems for early hand-related disorder detection.

How emotions are metaphorically embodied: measuring hand and head action strengths of typical emotional states

ABSTRACT This study measured hand and head action strengths of eight typical emotional states using an authentic but implicit emotion elicitation task. Participants listened to and then retold five stories in which eight typical emotional states were experienced by the narrators. The number of hand and head gestures that occur naturally while experiencing an emotional state was used as an index to determine the hand and head action strength of that emotional state. Results showed a larger number of head gestures than hand gestures, suggesting that head action strengths of the eight emotional states are stronger than their hand action strengths. These findings are consistent with the data extracted from Lancaster Sensorimotor Norms (LSN), although the two sets of data were gathered in two completely different experimental conditions and in two different languages. Furthermore, our data showed a prototypical directionality effect for the typical emotional states, specifically, happiness, anger and pride were primarily accompanied by upward gestures but downward gestures for sadness and shame; surprise was primarily accompanied by forward gestures but backward gestures for fear and disgust.

The effect of fatigue on climbing fluidity and hand movements

ABSTRACT In rock climbing, climbers use their arms to regulate their posture on the wall, which can lead to localised muscle fatigue. Evidence shows fatigue is the primary cause of falls, but little is known about how fatigue specifically affects climbing rhythm and hand movements. The present study examined climbing fluidity and hand movements on an indoor climbing wall before and after a specific fatiguing protocol. Seventeen climbers completed three repetitions of a challenging climbing route (21 on Ewbank scale) with different levels of localised arm fatigue. Climbers’ movements were tracked using 3D motion capture, and their hand actions assessed using notational analysis. Seventy markers were used to create 15 rigid body segments and the participants' centre of mass. The global entropy index was calculated on the path of the participants' centre of mass. Climbers fell more often when fatigued, but there were no significant differences in hip jerk or global entropy index when fatigued. No significant differences were found between the number of exploratory or performatory hand movements with different amounts of fatigue. The results suggest that localised arm fatigue affects a climber’s ability to prevent themselves from falling, but it does not specifically affect their fluidity.

Grid-Related Fine Action Segmentation Based on an STCNN-MCM Joint Algorithm during Smart Grid Training

Smart grid-training systems enable trainers to achieve the high safety standards required for power operation. Effective methods for the rational segmentation of continuous fine actions can improve smart grid-training systems, which is of great significance to sustainable power-grid operation and the personal safety of operators. In this paper, a joint algorithm of a spatio-temporal convolutional neural network and multidimensional cloud model (STCNN-MCM) is proposed to complete the segmentation of fine actions during power operation. Firstly, the spatio-temporal convolutional neural network (STCNN) is used to extract action features from the multi-sensor dataset of hand actions during power operation and to predict the next moment’s action to form a multi-outcome dataset; then, a multidimensional cloud model (MCM) is designed based on the motion features of the real power operation; finally, the corresponding probabilities are obtained from the distribution of the predicted data in the cloud model through the multi-outcome dataset for action-rsegmentation point determination. The results show that STCNN-MCM can choose the segmentation points of fine actions in power operation in a relatively efficient way, improve the accuracy of action division, and can be used to improve smart grid-training systems for the segmentation of continuous fine actions in power operation.

Morphology, Connectivity, and Encoding Features of Tactile and Motor Representations of the Fingers in the Human Precentral and Postcentral Gyrus.

Despite the tight coupling between sensory and motor processing for fine manipulation in humans, it is not yet totally clear which specific properties of the fingers are mapped in the precentral and postcentral gyrus. We used fMRI to compare the morphology, connectivity, and encoding of the motor and tactile finger representations (FRs) in the precentral and postcentral gyrus of 25 5-fingered participants (8 females). Multivoxel pattern and structural and functional connectivity analyses demonstrated the existence of distinct motor and tactile FRs within both the precentral and postcentral gyrus, integrating finger-specific motor and tactile information. Using representational similarity analysis, we found that the motor and tactile FRs in the sensorimotor cortex were described by the perceived structure of the hand better than by the actual hand anatomy or other functional models (finger kinematics, muscles synergies). We then studied a polydactyly individual (i.e., with a congenital 6-fingered hand) showing superior manipulation abilities and divergent anatomic-functional hand properties. The perceived hand model was still the best model for tactile representations in the precentral and postcentral gyrus, while finger kinematics better described motor representations in the precentral gyrus. We suggest that, under normal conditions (i.e., in subjects with a standard hand anatomy), the sensorimotor representations of the 5 fingers in humans converge toward a model of perceived hand anatomy, deviating from the real hand structure, as the best synthesis between functional and structural features of the hand.SIGNIFICANCE STATEMENT Distinct motor and tactile finger representations exist in both the precentral and postcentral gyrus, supported by a finger-specific pattern of anatomic and functional connectivity across modalities. At the representational level, finger representations reflect the perceived structure of the hand, which might result from an adapting process harmonizing (i.e., uniformizing) the encoding of hand function and structure in the precentral and postcentral gyrus. The same analyses performed in an extremely rare polydactyly subject showed that the emergence of such representational geometry is also found in neuromechanical variants with different hand anatomy and function. However, the harmonization process across the precentral and postcentral gyrus might not be possible because of divergent functional-structural properties of the hand and associated superior manipulation abilities.

Evaluation of Paste and Aqueous Lubricants on the Incidence of File Breakage, Quality of Obturation and Postoperative Pain in Children after Pulpectomy in Primary Teeth: A Randomised Clinical Trial

Introduction: Pulpectomy in primary teeth involves various steps, including tooth isolation, access opening, canal instrumentation, and obturation. Lubricants are commonly used in the biomechanical preparation of root canals to suspend and emulsify debris generated by the mechanical action of hand and rotary files. While paste-type lubricants (Prime Dental RC Help) and aqueous lubricants (17% aqueous Ethylenediaminetetraacetic acid [EDTA]) are available for use, there is a lack of studies in the literature documenting the effects of these lubricants on file fracture, obturation quality, and postoperative pain. Aim: To compare the effects of paste-type lubricants with aqueous lubricants on file breakage, quality of obturation, and postoperative pain in children treated with pulpectomy in primary teeth. Materials and Methods: This randomised clinical trial involved 40 primary teeth, evenly distributed for instrumentation using Kedo-S Plus pediatric rotary files. In Group 1 (n=20), paste-type lubricant was used during canal preparation, while in Group 2 (n=20), aqueous lubricant was used. The clinician recorded data on file fractures, evaluated the quality of obturation using immediate postoperative radiographs, and measured postoperative pain using the Four-point facial pain intensity rating scale. Descriptive statistics were performed once the data was tabulated. Results: Total of 40 primary teeth were included, participants aged 4-9 years, 22 girls and 18 boys. There was no incidence of file fracture in the group-2 treated with aqueous lubricants. Regarding the quality of obturation, the group-1 treated with paste-type lubricant had a higher number of under-fill and over-fill cases (p=0.002, 0.001), while no statistically significant difference was observed between the groups in terms of optimal fill (p=0.661). At 6, 12, 24, and 48 hours, there was no difference in pain experienced by the participants in both groups (p>0.05). Conclusion: During rotary instrumentation in primary teeth, aqueous lubricants were found to be more effective than pastetype lubricants in preventing file breakage, providing optimal obturation quality, and reducing postoperative pain.

Тембровые особенности «Сонаты с двух нот» для домры и фортепиано Е. М. Попляновой

This article examines the work of the Chelyabinsk composer Elena Mikhailovna Poplyanova. The main content of the study is the analysis of a new version of «Sonata with two notes» for domra and piano in the performed version of Maria Lebedeva. The article is based on the analysis of this work, determining the timbre features of the ensemble of domra and piano. The composer’s desire to feel the embodiment of musical images in a new way is emphasized, thanks to the use of complex rhythms, jumps, dynamic shades and the use of all registers in the solo parts of the instruments. One of the important elements of the construction of the sonata is the individual properties of the creative thinking of the composer, which reveals the artistic image of lyrical and grotesque characters. The article emphasizes the peculiarity of the rhythmic stability, where specific articulatory expressive means become a support. The main aspects revealed are the characteristic features of the sonata genre, the features of concerto, the content, and the compositional structure. Most important is the performance: what tasks are set for the musicians, the coordination of hand actions and sound extraction on the domra come to the fore. The performance of all musical techniques becomes one of the main ways to convey the figurative content of the «Sonata from Two Notes».

Learning Dynamic Robot-to-Robot Object Handover

Object handover is an essential skill for collaborative robots in both services robotics and manufacturing scenarios. Most previous works were conducted from the perspective of human-robot interaction. The object handover between robots for collaborative task execution, aiming at optimizing time efficiency with human-like smooth behaviour, has not been extensively addressed. In this work, we propose a skill framework based on variable impedance control and dynamic motion primitives to optimize not only the motion trajectories and variable impedance, but also the timing of hand actions during dynamic motion. The effectiveness of the proposed framework is evaluated on a real dual-arm robot system under two handover scenarios with different constraints on the timing of hand actions. The experiment results demonstrate significant time efficiency improvement with reduction of the execution time by 9.4% and 23.7%, compared with accelerating the motion speed of the demonstrated handover. Furthermore, it can be observed that the robot successfully learned dynamic object handover without requiring transfer action to be triggered after both hands stop and remain still. In addition, in the second experiment, it is shown that the object can be transferred even without ensuring firm contact, which indicates that object handover is possible to be realized by throwing-like motion.

Hand action recognition in rehabilitation exercise method using R(2+1)D deep learning network and interactive object information

Hand action recognition in rehabilitation exercises is to automatically recognize what exercises the patient has done. This is an important step in an AI system to assist doctors to handle, monitor and assess the patient’s rehabilitation. The expected system uses videos obtained from the patient's body-worn camera to recognize hand action automatically. In this paper, we propose a model to recognize the patient's hand action in rehabilitation exercises, which is a combination of the results of a deep learning network recognizing actions on Video RGB, R(2+1)D, and a main interactive object in the exercises detection algorithm. The proposed model is implemented, trained, and tested on a dataset of rehabilitation exercises collected from wearable cameras of patients. The experimental results show that the accuracy in exercise recognition is practicable, averaging 88.43% on the test data independent of the training data. The action recognition results of the proposed method outperform the results of a single R(2+1)D network. Furthermore, the better results show the reduced rate of confusion between exercises with similar hand gestures. They also prove that the combination of interactive object information and the action recognition improve the accuracy significantly.

Обґрунтування параметрів контролю спеціальної фізичної підготовленості спортсменів віком 12–13 років у таеквон-до ІТФ

Physical preparedness is a basic type of athletes’ preparedness in all versions of taekwon-do. Attention to physical preparation remains fixed and actual at various stages of multi-year sports improvement. Researchers and methodological papers mostly present specifics of training and competition activity in taekwon-do WTF and to a lesser extent taekwon-do ITF and GTF. This calls for specification of the set of information means for control of specialized physical preparedness of athletes aged 12–13. The aim of the paper is to substantiate parameters of control of specialized physical preparedness of athletes aged 12–13 in taekwon-do ITF. Materials and methods. The following methods have been used: the method of theoretical analysis and academic and methodological references data generalization; analysis of documented materials; the method of expert polling; methods of mathematical and statistical data processing. The structure and content of Training programs in taekwon-do ITF, WTF, rules of competitions in taekwon-do ITF have been studied. 24 coaches in taekwon-do ITF were involved in expert polling; presentation group of young athletes aged 12-14 (15 persons) whose training experience in taekwon-do ITF was over 5 years were additionally involved. The questions were related to determination of time spells priority for control of athletes’ special speed qualities; time spells for control of athletes’ special power endurance; kicks and punches as well as combination of two-three hits that can be used for control of athletes’ special physical preparedness. Polling results processing was done via ranking of the suggested variants by shares of respondents’ answers. Results.It has been revealed that when elaborating the program of testing of special physical preparedness of athletes aged 12–13 in taekwon-do ITF it is reasonable to rely on the use of actions in two modes: speed and power mode (acting for 10-30 seconds), and speed and power endurance (90–120 seconds). Execution of single punches in control of physical preparedness must include hand actions: jab, upper-cut or cross, and kicks: Dollyo Chagi, Yop Chagi, or Naeryo Chagi. Execution of combination of two hits, according to expert recommendations, should be done in: “Dollyo Chagi-cross” and “Yop Chagi-hook”, while combination of three hits – “hook-Dollyo Chagi-dwi chagi” and “jab – Yop Chagi-hook”. Conclusions. Parameters of control have been substantiated for the first time; use of single kicks and punches, execution of two and three hits for control of special physical preparedness of athletes aged 12–13 in taekwon-do ITF have been specified

A CW-CNN regression model-based real-time system for virtual hand control.

For upper limb amputees, wearing a myoelectric prosthetic hand is the only way for them to continue normal life. Even until now, the proposal of a high-precision and natural performance real-time control system based on surface electromyography (sEMG) signals is still challenging. Researchers have proposed many strategies for motion classification or regression prediction tasks based on sEMG signals. However, most of them have been limited to offline analysis only. There are even few papers on real-time control based on deep learning models, almost all of which are about motion classification. Rare studies tried to use deep learning-based regression models in real-time control systems for multi-joint angle estimation via sEMG signals. This paper proposed a CW-CNN regression model-based real-time control system for virtual hand control. We designed an Adaptive Kalman Filter to smooth the joint angles output before sending them as control commands to control a virtual hand. Eight healthy participants were invited, and three sessions experiments were conducted on two different days for all of them. During the real-time experiment, we analyzed the joint angles estimation accuracy and computational latency. Moreover, target achievement control (TAC) test was applied to emphasize motion regression in real-time. The experimental results show that the proposed control system has high precision for 3-DOFs motion regression in simultaneously, and the system remains stable and low computational latency. In the future, the proposed real-time control system can be applied to actual prosthetic hand.

CAFI-AR

Freehand interaction enhances user experience, allowing one to use bare hands to manipulate virtual objects in AR. Yet, it remains challenging to accurately and efficiently detect contacts between real hand and virtual object, due to the imprecise captured/estimated hand geometry. This paper presents CAFI-AR, a new approach for Contact-Aware Freehand Interaction with virtual AR objects, enabling us to automatically detect hand-object contacts in real-time with low latency. Specifically, we formulate a compact deep architecture to efficiently learn to predict hand action and contact moment from sequences of captured RGB images relative to the 3D virtual object. To train the architecture for detecting contacts on AR objects, we build a new dataset with 4,008 frame sequences, each with annotated hand-object interaction information. Further, we integrate CAFI-AR into our prototyping AR system and develop various interactive scenarios, demonstrating fine-grained contact-aware interactions on a rich variety of virtual AR objects, which cannot be achieved by existing AR interaction approaches. Lastly, we also evaluate CAFI-AR, quantitatively and qualitatively, through two user studies to demonstrate its effectiveness in terms of accurately detecting the hand-object contacts and promoting fluid freehand interactions