Source-linked AI summary
Intelligent Approaches to interact with Machines using Hand Gesture Recognition in Natural way: A Survey
Ankit Chaudhary, J. L. Raheja, Karen Das, Sonia Raheja
TL;DR
Hand gesture recognition offers natural, device-free human–machine interaction but remains difficult because hands and gestures vary substantially. This survey organizes computer-vision work around intelligent and soft-computing methods, image preprocessing, and hand-image construction, finding fingertip detection to be a common appearance-based strategy while identifying limited work on individual finger bending and movement.
Problem
Hand gesture recognition must support natural human–machine interaction despite hand deformability, gesture diversity, ambiguity, and temporal and spatial variation.
Method
The paper surveys computer-vision approaches using artificial neural networks, fuzzy logic, genetic algorithms, image preprocessing, segmentation, and hand-image construction.
Results
The survey finds that fingertip detection is a primary focus in appearance-based hand recognition and compares intelligent approaches and reported results across prior work.
Takeaways & Limitations
Soft computing is presented as useful when exact hand or finger positions cannot be determined, while fingertip-based methods dominate reported appearance-based work.
Takeaways & Limitations
Few existing studies address individual finger-position bending and movements, with most focusing on full-hand or fingertip positions.
Abstract
from arXiv · showhide
Hand gestures recognition (HGR) is one of the main areas of research for the engineers, scientists and bioinformatics. HGR is the natural way of Human Machine interaction and today many researchers in the academia and industry are working on different application to make interactions more easy, natural and convenient without wearing any extra device. HGR can be applied from games control to vision enabled robot control, from virtual reality to smart home systems. In this paper we are discussing work done in the area of hand gesture recognition where focus is on the intelligent approaches including soft computing based methods like artificial neural network, fuzzy logic, genetic algorithms etc. The methods in the preprocessing of image for segmentation and hand image construction also taken into study. Most researchers used fingertips for hand detection in appearance based modeling. Finally the comparison of results given by different researchers is also presented.
1 Introduction
The introduction presents natural hand gesture recognition as a device-free human–machine interaction area, while organizing prior approaches by how they process and represent hand images. It highlights convenience alongside challenges from hand complexity, gesture variation, and computational cost.
- Motivation: Natural hand gesture recognition enables device-free interaction with machines through ordinary hand movements.The paper connects this interaction mode to applications including sign language and human–machine interfaces.
- Prior Systems: Researchers have used gloves, sensors, wires, color strips, and shirts to measure or detect hand movements, although the survey focuses on natural vision-based interaction.The introduction contrasts these instrumented techniques with non-contact gesture recognition.
- Approach Classification: The survey divides approaches into appearance-based methods that reconstruct hand images and model-based methods that represent images using hand models.This classification is based on the method used rather than how the image is treated.
- Challenges: Gesture recognition is difficult because hands are deformable and gestures exhibit diversity, ambiguity, and temporal and spatial differences.The paper also characterizes human vision as an ill-posed problem in this context.
- Computational Considerations: Algorithm simplification and reduced processing time are identified as important concerns in gesture-recognition systems.The introduction notes parallel implementations as one response to processing demands.
2 Hand Detection Approaches
Hand detection approaches combine preprocessing, skin or background segmentation, fingertip localization, and geometric or model-based reconstruction. Fingertips are a prominent basis for constructing hand images, while systems also use color models, morphology, filters, and articulated hand representations.
- Overview: Hand-detection techniques are divided into two broad approach groups before describing specific preprocessing and reconstruction methods.The section subsequently discusses fingertip-based, appearance-based, and model-based systems.
- Fingertip-Based Detection: Many researchers use fingertip detection to construct hand images, treating fingertips as natural determinants of hand posture.Reported techniques include colored fingertip histograms, prototype templates, corner detection, morphology, and geometric calculations.
- Segmentation: Skin segmentation and color modeling support hand-region detection, using Gaussian, RGB, HSI, and related color representations.Background subtraction and histogram-based skin-color probabilities are also used to acquire gesture images.
- Image Processing: Researchers apply morphology, Gaussian filters, Freeman’s algorithm, Kalman filtering, and Markov Random Fields for localization, noise removal, or image refinement.These methods address fingertip positioning, perimeter extraction, or noise components in processed images.
- Model-Based Approaches: Model-based recognition represents finger and thumb joints with degrees of freedom and local coordinate systems to model interdependent finger movement.One described system uses fast fitting with joint-centered coordinate frames.
3 Soft Computing Approaches
The survey organizes soft-computing approaches for hand gesture recognition around neural networks, fuzzy logic, genetic algorithms, and related learning or image-processing methods. These approaches support fingertip, trajectory, pose, classification, and gesture-extraction tasks.
- Soft computing approaches discussed include artificial neural networks, fuzzy logic, genetic algorithms, machine learning, evolutionary computation, and probabilistic reasoning.
- 3.1 Artificial Neural Network: ANN-based methods learn hand features such as fingertips and transform fingertip vectors into finger-joint angles for articulated hand models.
- 3.1 Artificial Neural Network: HMM and RNN classifiers can combine outputs to classify gestures from vectors representing hand-pose frames.
- 3.1 Artificial Neural Network: SGONG uses a self-growing, self-organized neural network to recognize predefined gestures from raised-finger configurations.The approach identifies a fingertip neuron, follows a neuron chain, estimates the palm center, and applies likelihood classification.
- 3.2 Fuzzy Logic Based Approaches: Fuzzy approaches use weighted cluster centroids, map clusters to finite-state-machine states, and can represent gesture positions linguistically.One described finite-state-machine approach was not implemented, while another applied fuzzy logic to selected sitting and standing positions.
- 3.4 Other Approaches: Other approaches apply PCA, SVM, entropy measures, coordinate transformations, and dynamic space-time warping for extraction, dimensionality reduction, and recognition.The survey also describes video processing, hand coordinates, eigenvectors, and entropy-based image analysis as parts of gesture-recognition systems.
4 Implementation Tools
The surveyed implementations commonly use MATLAB and image-processing tools, with some systems using C++, OpenCV, Delphi, or platform-specific interfaces. Reported recognition accuracy varies with gesture type, lighting, image quality, and the selected method.
- Most image-processing systems use MATLAB with its image-processing toolbox, while some implementations use C++, OpenCV, or Delphi.C++ systems described in the survey were implemented on Windows XP, and one SGONG system used Delphi.
- 90–95% accuracy was reported for open fingers, compared with 10–20% for closed fingers in one morphology-based extraction system.The lower closed-finger accuracy was attributed to image quality and morphology-operator limitations when fingers appear in front of the palm.
- Reported results include 90% under good lighting, 90.45% for an SGONG system, 91.9% using HMM and RNNs, and 93.7% using Gabor filters.
6 Conclusions
The survey examines natural hand gesture recognition through intelligent and soft-computing approaches, emphasizing applications, image-processing methods, and fingertip-based detection. It concludes that soft computing is useful when exact hand or finger positions cannot be determined, while individual finger movement remains underexplored.
- The survey covers natural hand gesture recognition applications and intelligent approaches including artificial neural networks, fuzzy logic, and genetic algorithms.
- Soft computing supports recognition when exact hand or finger positions are unavailable by using approximation, learning models, and training data.
- Most researchers addressed full-hand or fingertip positions, leaving individual finger bending and movement as areas for future work.