|
| Study | Year | Dataset | Remarks |
|
| (a) |
|
| [141] | 2021 | PSL dataset | 37 alphabets |
|
| [165] | 2021 | ISLAN (Indian Sign Language) | Collection of 700 sign images, and 24 sign videos |
|
| [139] | 2021 | SIBI dataset | 8 static word signs. 19200 total images are included |
|
| [140] | 2021 | | Custom made numbers from 1 to 5 |
|
| [142] | 2021 | RKS persiansign, first-person, ASVID, isoGD | (i) RKS-PERSIANSIGN: this dataset comprises 10,000 RGB videos showcasing 100 Persian sign words. These videos are contributed by 10 individuals, including 5 women and 5 men, with 100 video samples available for each Persian sign word |
| (ii) First-person: this dataset consists of 100,000 RGB-D frames depicting 45 different hand action categories performed with 26 distinct objects, capturing various hand configurations. Only the RGB sequences from the ASVID dataset are used in this context |
| (iii) isoGD: this dataset contains a total of 47,933 RGB and depth video samples across 249 class labels. For your reference, only the RGB samples are utilized in this dataset. It is further divided into three subdatasets, with 35,878 samples designated for training, 5,784 samples for validation, and 6,271 samples for testing |
|
| [137] | 2020 | HamNoSys database | 3000 words |
|
| [138] | 2020 | Chinese Sign Language | The dataset generated consists of 51 common word signs from which 60 sentences were created. Instances of sentences are 20400 from 34 volunteers |
|
| [127] | 2020 | Korean Sign Language | 17 words used for training |
|
| [128] | 2020 | China Sign Language | Data augmentation is used to obtain a benchmark dataset based on Chinese Sign Language (CSL). One dataset is obtained from Kaggle and the other is built from 30-second video frames |
|
| [120] | 2020 | American Sign Language (ASL) and Bengali Sign Language (BdSL) | A dataset is generated which contains 1000 data points for each of the letters of ASL and BdSL |
|
| [132] | 2020 | MS-ASL dataset | This dataset has 25000 clips over 222 signers and covers 1000 most frequently used ASL gestures |
|
| [133] | 2020 | Bangla Sign Language | This dataset has 30 consonants and 6 vowels of BSL characters. The dataset holds 36 × 50 = 1800 images in total as it has 50 samples for each sign |
|
| [129] | 2020 | German Sign Language | The dataset has 301 videos with an average duration of 9 minutes |
| [125] | 2020 | American Sign Language | A dataset consisting of 80 video clips that focus on finger movement. These video clips were sourced from two different origins: 32 were extracted from publicly available videos, while the remaining 48 clips were recorded manually. Within this dataset, there are 20 instances for each of the four alphabets: D, I, J, and Z |
|
| [131] | 2020 | Croatian Sign Language | The dataset was generated which consists of 25 languages and their signs. 40 volunteers performed each gesture twice which resulted in 2000 sign videos |
|
| [122] | 2020 | Hong Kong Sign Language (HKSL) | The dataset was created by the authors. It consists of 45 most common words. For each word, 30 videos from different signers were recorded. Total videos are 1500 |
|
| [134] | 2020 | Indian Sign Language | Custom created. The dataset includes 100 static signs, that is, 23 English alphabets, 0–10 digits, and 67 commonly used words. There are 300 images of each instance totaling 35000 images |
|
| [135] | 2020 | Custom made | The dataset has four unique word signs. Each sign has 50 images with different positions and light levels. The total number of images is 1000 |
|
| [123] | 2020 | German Sign Language | Public dataset. RWTH-PHOENIX-weather 2014 |
|
| [136] | 2020 | RKS-PERSIANSIGN first-person dataset NYU hand pose dataset | (i) RKS-PERSIANSIGN: |
| (1) Contains: 10,000 RGB videos |
| (2) Content: 100 Persian words |
| (3) Contributors: 10 individuals |
| (4) Purpose: likely used for Persian sign language recognition. This dataset provides video samples for training and evaluating models for recognizing |
| (ii) Persian sign language gestures |
| First-person dataset: |
| (1) Contains: 100,000 RGB-D frames |
| (2) Content: 45 hand action categories for 26 different objects |
| (3) Purpose: this dataset seems focused on recognizing hand actions related to interactions with various objects. The RGB-D frames can be used for training and evaluating models capable of understanding hand-object interactions |
| (iii) NYU hand pose dataset: |
| (1) Contains: 81,009 image sequences |
| (2) Content: 36 joints |
| (3) Purpose: likely used for hand pose estimation. This dataset provides a large number of image sequences capturing various hand poses, which can be used to train and test models for hand pose estimation tasks |
| [126] | 2020 | Flemish Sign Language | Public dataset |
| The total samples are 18730 from 67 native signers with 100 classes |
|
| [102] | 2019 | The dataset contains three gestures | The three gestures are feeling uncomfortable, seeing a doctor, and taking medicine |
|
| [110] | 2019 | ASL alphabet dataset. Sign language and static gesture recognition dataset | (i) The ASL alphabet dataset contains 87,000 images. The sign language and static gesture recognition dataset contains 1,687 images |
| (ii) The authors created their dataset from these two datasets which contain 73,488 images |
|
| [105] | 2019 | American Sign Language | A total of 10 samples of each alphabet were taken for accuracy |
|
| [103] | 2019 | Arabic Sign Language | 10 alphabets Alif, Ba, Ta, Kha, Dal, Dhad, Thah, Ghayn, Lam, and La. 2000 images used for training |
|
| [104] | 2019 | British Sign Language | 26 letters A to Z |
| Training performed on 520 samples (26 classes with 20 samples per class) |
|
| [109] | 2019 | Indonesian language inflectional words | Custom made |
| (i) Word count: the dataset consists of a total of 1,440 inflectional words |
| (1) Training data: 954 inflectional words |
| (2) Testing data: 486 inflectional words |
| (ii) Data sources: the data were recorded by three teachers from Santi Rama school for the hearing impaired in Jakarta |
|
| [91] | 2019 | ASL dataset | Two datasets: one is word-level (70 ASL words) and the other is sentence-level (100 sentences) |
|
| [101] | 2019 | Arabic Sign Language | Only 5 letters were taken for the experiment |
|
| [90] | 2019 | Custom-made | 5 volunteers to perform 26 alphabet signs with 30 repetitions. That is, 26 × 30 × 5 alphabet signs (3,900) in the dataset |
|
| [87] | 2019 | Swedish Sign Language signs dataset | Swedish keyword signing targeted children with communicative disorders |
|
| [115] | 2019 | Custom-made | 40 signs five times each totaling 200 for testing |
| [116] | 2019 | Custom-made PSL | (i) Dataset generation: the dataset was generated by capturing videos of sign language gestures. Afterward, frames were extracted from these videos using the Matlab image processing toolbox |
| (ii) Signs: the dataset includes various sign language gestures, with each sign represented by a substantial number of pictures |
| (iii) Number of signs: not specified, but there are multiple signs |
| (iv) Pictures per sign: each sign is represented by approximately 1,500 to 2,000 pictures |
| (v) Total pictures: the dataset contains a total of around 21,000 pictures |
|
| [99] | 2019 | German Sign Language weather forecast program | RWTH-PHOENIX-weather-2014 |
| (i) Training set: |
| (1) Number of videos: 5,672 |
| (2) Use: typically used to train machine learning or deep learning models |
| (ii) Validation set: |
| (1) Number of videos: 540 |
| (2) Use: used during the model development process to fine-tune hyperparameters and assess model performance |
| (iii) Test set: |
| (1) Number of videos: 629 |
| (2) Use: reserved for evaluating the final model’s performance and assessing its generalization to unseen data |
|
| [97] | 2019 | Ghanaian Sign Language | Custom-made |
| The dataset consists of 66000 images in RGB color with 33 classes of static gestures having 24 alphabets and 9 digits |
|
| [94] | 2019 | Korean Sign Language | Custom-made |
| Ten words were selected. A different number of videos were selected from the Internet for each word. The total no. of videos is 421 |
|
| [93] | 2019 | CSL | The authors selected 100 kinds of sign language words. The training set consists of 2964, the validation set has 1044, and the testing set has 1005 videos |
|
| [119] | 2019 | ASL | Custom-made |
| 26 alphabets |
| [86] | 2019 | ASL Russian Sign Language (RSL) | ASL dataset: Massey University of researchers |
| This dataset consists of 2425 images from 5 individuals |
| RSL: |
| Custom-made |
| The data for RSL are collected from five YouTube videos. The total number of gestures in RSL is 33. Only the 26 static gestures are taken and the rest of the dynamic gestures are not included in this work |
|
| [95] | 2019 | Custom made | The static sign language has 24 alphabets. J and Z are excluded because they are dynamic. Also, it included and captured from seven native and nonnative signers with alike lighting |
|
| [106] | 2019 | ASL | There are 6000 words in the ASL dictionary |
|
| [117] | 2019 | ASL | Public dataset |
| The dataset collected from Kaggle contains pictures of static hand motions of ASL with 24 classes. The database consists of 47475 pictures from which 33000 (70%) pictures were used in the training set and 1445 (30%) pictures for testing |
|
| [114] | 2019 | LSA64 dataset | Public dataset: |
| Argentinian Sign Language | The authors selected 30 gestures and 50 video streams for each gesture. After video processing, 90,000 images were created representing the sequence of dynamic gestures. The number of images for each category is 3000 |
|
| [6, 118] | 2019 | ASL | A comprehensive collection of American Sign Language (ASL) gestures representing 24 English letters (excluding “Y” and “Z”). These gestures are captured in the form of expressive hand movements, providing a rich resource for ASL recognition |
| These ASL gestures used Kinect technology with contributions from 5 different individuals |
|
| [100] | 2019 | ASL | Public dataset |
| ASLLVD, the American Sign Language lexicon video dataset, features nearly 10,000 ASL signs by 6 native signers. The dataset focuses on 50 hand-picked ASL signs, each signed by 6 different individuals, totaling 300 videos. These videos include various angles, but our analysis concentrated on front-view recordings |
|
| [96] | 2019 | ASL | Custom-made |
| The authors collected video data for 25 ASL signs from 100 users where each sign was executed three times each. The total number of instances was 7500 |
| [107] | 2019 | ISL | Custom-made |
| The authors selected 26 common signs. Each sign sample comprises 50 consecutive readings, representing 50-time points of gesture motion. A single sample is structured as a 50 × 11 matrix, forming 2D data stored in a CSV file |
|
| [108] | 2019 | SIBI | Custom-made |
| The number of videos is 2275 which consists of 28 common sentences |
|
| [98] | 2019 | ASL | Custom-made |
| 26 letters of the ASL alphabet are included. The signers are 3 and each signer took 10 signs for each alphabet which totals 30 for one alphabet. Thus, the total number of instances is 780 |
|
| [113] | 2019 | ISL | The dataset consists of 2500 images for alphabets and dynamic words. The authors augmented this dataset and produced 5157 images |
|
| [88] | 2019 | CSL | The authors have created a database of four tables to store symbols with important descriptions. They have used HamNoSys which consists of 200 symbols consisting of hand shapes, hand orientation, location, and movements |
| ASL |
|
| [112] | 2019 | ASL | Custom-made |
| The study concentrates on static ASL gestures from A to Y, omitting J and Z due to their dynamic nature. The dataset comprises 24 gesture images captured with a smartphone camera. Each gesture is represented by 200 images taken by two users, accounting for a total of 4800 images |
|
| [92] | 2019 | Thai Sign Language (TSL) | Custom-made |
| The authors used Microsoft Kinect to record the video stream dataset. It consists of 64 isolated vocabularies. Each word was performed by 8 nonnative TSL signers and each signer acted 5 times for each word. Thus, there are 64 × 8 × 5 = 2560 video samples in total |
|
| [89] | 2019 | Brazilian Sign Language | Custom-made |
| Authors recorded videos for 26 letters of the alphabet in Libras with 13 users. The total number of videos was 338 |
|
| [74] | 2018 | Indonesian Sign Language | Alphabets A to Z and numbers 1 to 10 used in this experiment |
|
| [70] | 2018 | Indonesian Sign Language | Alphabets A to Z taken |
|
| [84] | 2018 | The open dataset given at Kaggle called sign language MNIST | A set of 28 × 28 images representing the standard American Sign Language (ASL) alphabet, excluding J and Z |
| [82] | 2018 | French Sign Language | 22 gestures were taken out of 26 from French Sign Language. 4 gestures, that is, J, P, Y, and Z, were left out because of their nonstatic nature. Each gesture was performed by 57 participants. The total dataset contains 1.25 million samples |
|
| [75] | 2018 | Indian Sign Language (ISL) | Digits 0 to 9 and alphabets a to z were taken for the experiment |
|
| [79] | 2018 | Indian Sign Language (ISL) | Digits 0 to 9 and alphabets a to z were taken for the experiment |
|
| [83] | 2018 | Custom built. Indian Sign Language | 18 signs with each sign by 10 different signers recorded |
|
| [71] | 2018 | Indian Sign Language
American Sign Language
British Sign Language
Turkish Sign Language | (i) ISL dataset: used SVM for this dataset |
| Contains 4 signs, that is, A, B, C, and the word “Hello” |
| (ii) ASL dataset: used KNN for this dataset |
| Contains 10 ASL fingerspelling alphabets from a to i and k. The letter j is not included. The total number of samples was 5254 |
| (iii) ISL: used CNN for this dataset |
| The total dataset is 5000 samples for 200 signs done by five Indian Sign Language users |
| (iv) Authors used ANN for the following 3 datasets |
| (v) ASL: consists of letters from A to Z |
| (vi) British Sign Language: contains alphabets from A to Z |
| (vii) Turkish Sign Language: |
| Consists of alphabets from A to Z. The letters Q, W, and X are excluded |
|
| [72] | 2018 | Argentinian Sign Language | LSA64 dataset: 10 subjects, 5 repetitions, 64 sign types, 3200 videos |
| RWTH-PHOENIX-weather database: 50 classes, 1297 training videos, 238 testing videos |
|
| [73] | 2018 | | Public dataset |
| There are 900 pictures including 25 samples for each of 36 characters consisting of 26 letters and 10 digits |
|
| [77] | 2018 | ISL | Custom-made |
| 200 sign language words. Each sign is performed by 5 different signers |
|
| [80] | 2018 | ISL | Custom-made |
| A dataset of 5000 images and 100 images each for 50 most commonly used words was created |
|
| [85] | 2018 | ISL | Custom-made |
| The dataset consists of 200 words to form sentences |
| [81] | 2018 | ASL | Massey University gesture dataset 2012: |
| Consists of 36 classes with 2524 images |
| ASL fingerspelling A dataset: |
| Consists of 24 classes with 131,000 images |
| NYU: |
| Consists of 36 classes with 81,009 images |
| ASL fingerspelling dataset of the Surrey University: |
| Consists of 24 classes with 130,000 images |
|
| [78] | 2018 | ASL | ASL alphabet dataset: public dataset |
| There are 24 static gestures from letters A–Y. J is excluded as it is dynamic. There are 100 images for each class |
|
| [69] | 2018 | Korean Sign Language | Custom-made |
| The dataset consists of 10,480 videos collected from ten Korean professional signers |
|
| [76] | 2018 | SIBI (Sistem Isyarat Bahasa Indonesia) | Custom-made |
| The dataset consists of SIBI words performed by 2 teachers fluent in this language. It consists of 21 root words and 155 inflectional words. Each word is recorded 5 times by each teacher, thus resulting in a total of 1760 signs |
|
| [60] | 2017 | Custom-made | Static gestures for the English alphabets from A to Z and digits from 0 to 9 |
|
| [62] | 2017 | Custom-made | Static gestures for the English alphabets from A to Z and digits from 0 to 9 |
|
| [2] | 2017 | Custom-made | Static gestures for the English alphabets from A to Z and digits from 0 to 9 |
|
| [67] | 2017 | Indonesian Sign Language | Dataset: 1000 samples, 50 Indonesian sign words, 20 samples per sign, 500 for training, 500 for testing |
|
| [61] | 2017 | ISL | Custom-made |
| 26 alphabets from A to Z and 12 basic words |
|
| [66] | 2017 | ISL | Custom-made |
| 18 different words were included in the dataset |
|
| [56] | 2017 | Ubicomp.eti.uni | 18 different words were included in the dataset |
|
| [57] | 2017 | ASL | 103 signs |
|
| [68] | 2017 | ASL | The dataset has a total of 720 images (30 for every ASL sign image). The dataset consists of alphabets from A to Y. The letters J and Z are excluded |
|
| [64] | 2017 | Sinhala Sign Language (SSL) | Custom-made |
| The dataset consists of 61 SSL fingerspelling signs (words) and 40 SSL number signs |
| [58] | 2017 | Greek Sign Language | Custom-made |
| 5 participants (2 male, 3 female) learned and performed 15 signs, four times each, totaling 300 evaluation samples |
|
| [63] | 2017 | Korean Sign Language | Custom-made |
| 30 different gestures are included in this dataset. The training data are 72% and the testing data are 28% |
|
| [59] | 2017 | Thai Sign Language (TSL) | The dataset consists of 10 words. Each word has 10 samples |
|
| [55] | 2017 | NGT (Nederlandse Gebarentaal) sign language of the Netherlands | Public dataset |
| The dataset consists of 40 glosses (words) taken from the NGT dataset |
|
| [65] | 2017 | ASL | Custom-made |
| The dataset consists of 25 images from 5 people for each alphabet and digits 0–9 |
|
| [50] | 2016 | ASL | 16 alphabets taken for training and testing |
|
| [49] | 2016 | Indonesian Sign Language | 24 gestures from A to Y excluding J and Z |
|
| [45] | 2016 | ASL | Custom-made |
| The dataset consists of 20 ASL signs |
|
| [51] | 2016 | Arabic Sign Language (ArSL) | Public dataset: |
| This dataset consists of 588 signs which include 10 numbers from 0 to 9, 28 alphabets, and different categories like family, job, colors, and sports |
|
| [3] | 2016 | Pakistan Sign Language | 6 alphabets from A to F with 20 samples for each letter collected |
|
| [52] | 2016 | Continuous sign language | 18 signs with each sign by 10 different signers recorded |
|
| [48] | 2016 | Danish Sign Language | (i) Danish Sign Language: this dataset consists of 2,149 signs |
| New Zealand Sign Language | (ii) New Zealand Sign Language: this dataset consists of 4,155 signs |
| RWTH-PHOENIX-weather 2014 | (iii) RWTH-PHOENIX-weather 2014: this dataset consists of 65,227 signs |
|
| [47] | 2016 | German Sign Language | RWTH-PHOENIX-weather 2012 |
| RWTH-PHOENIX-weather multisigner 2014 |
| This dataset consists of 65,227 signs |
| SIGNUM single signer: |
| This dataset consists of 450 basic signs. Isolated signs are 450 and continuous sentences are 780. The total number of images is 5,970,450 |
|
| [44] | 2016 | American Sign Language image dataset (ASLID) | Public datasets |
| American Sign Language lexicon video dataset (ASLLVD) | Training set: 808 ASLID images from six signers. Test set: 479 ASLID images from two signers |
| [46] | 2016 | Greek Sign Language | Custom-made |
| 24 Greek Sign Language letters, 10 samples each, 6 subjects, totaling 1440 samples |
|
| [54] | 2016 | Korean Sign Language | Custom-made |
| Experiment: 5 subjects, 1–9 numbers repeated 5 times. 3 males and 2 females were the participants |
|
| [41] | 2015 | South African Sign Language (SASL) | Taken only three alphabets A, B, and C and three digits 1, 2, and 3 |
|
| [42] | 2015 | Malaysian Sign Language | Taken only three alphabets A, B, and C and three digits 1, 2, and 3 |
|
| [39] | 2015 | Taiwan Sign Language | 51 fundamental postures in Taiwan Sign Language |
|
| [35] | 2015 | ASL | Custom built (real-time hand gesture recognition system) |
|
| [37] | 2015 | Indonesian Sign Language | Alphabets A to Z |
|
| [40] | 2015 | ASL | Only the letters A to Z are included for testing |
|
| [43] | 2015 | Greek Sign Language | Greek Sign Language alphabets |
|
| [36] | 2015 | German Sign Language (DGS) | Public dataset |
| RWTH-PHOENIX-weather corpus: |
| Dataset: 2137 sentence segments, 14717 gloss annotations, 189,363 frames |
|
| [28] | 2014 | Custom built | Hand gesture image database |
| The test dataset was prepared by four persons each of whom showed 19 signs with three rotation variations |
|
| [33] | 2014 | PSL | 300 samples taken from 30 individuals with 10 signs each |
|
| [34] | 2014 | PSL | Custom-made |
| This dataset consists of 500 images of 37 alphabets. 426 images were utilized for training and 74 for testing |
|
| [31] | 2014 | Dataset DS1 | The number of one-handed videos and frames is 42 and 902 |
| The number of two-handed videos and frames is 48 and 1337 |
| Dataset DS2: |
| The number of one-handed videos and frames is 42 and 1276 |
| The number of two-handed videos and frames is 48 and 1945 |
| Dataset DS3: |
| The number of one-handed videos and frames is 42 and 1197 |
| The number of two-handed videos and frames is 48 and 1735 |
| [29] | 2014 | PSL | Custom-made |
| The dataset consists of 37 alphabets. 6 samples are recorded for each alphabet |
|
| [30] | 2014 | ASL | Custom-made dataset |
| Dataset: 24 static letters signed by 5 individuals, 60,000 images |
|
| [32] | 2014 | ArSL | The sign-to-letter translation by using a hand glove, microcontroller, and display unit |
|
| [27] | 2013 | Thai Sign Language (TSL) | Custom-made |
| The dataset consists of 42 TSL alphabets. Several videos are taken for each alphabet |
|
| [26] | 2012 | | Custom-built |
| A word is an input to the smartphone which is converted to video animation |
|
| [25] | 2012 | Brazilian Sign Language (Libras) | Custom-made |
| The dataset consists of two sets. One is the vowel set which is A, E, I, O, and U. The other set has the set which has B, C, F, L, and V |
|
| Name | Link (access date 25-August-2023) | | |
|
| (b) |
|
| PSL | https://data.mendeley.com/datasets/y9svrbh27n/1 | | |
|
| First-person | https://guiggh.github.io/publications/first-person-hands/ | | |
|
| Purdue RVL-SLLL | https://engineering.purdue.edu/RVL/Database/ASL/asl-database-front.htm | | |
|
| Corpus NGT | https://www.ru.nl/en/cls/research | | |
|
| isoGD | http://www.cbsr.ia.ac.cn/users/jwan/database/isogd.html | | |
|
| SIGNUM | https://www.phonetik.uni-muenchen.de/forschung/Bas/SIGNUM/ | | |
|
| WLASL | https://dxli94.github.io/WLASL/ | | |
|
| ASLID | http://vlm1.uta.edu/~srujana/ASLID/ASL_Image_Dataset.html | | |
|
| German Sign Language | https://www-i6.informatik.rwth-aachen.de/~koller/RWTH-PHOENIX/ | | |
|
| Danish Sign Language | https://www.tegnsprog.dk/ | | |
|
| ArSL | https://menasy.com/ | | |
| How2Sign | https://how2sign.github.io/ | | |
|
| GSL dataset | https://vcl.iti.gr/dataset/gsl/ | | |
|
| AUTSL | https://chalearnlap.cvc.uab.cat/dataset/40/description/ | | |
|
| LSA64 | https://facundoq.github.io/datasets/lsa64/ | | |
|
| Ubicomp | https://ubicomp.eti.uni-siegen.de/home/datasets/ | | |
|
| ASL finger spelling | https://www.kaggle.com/datasets/mrgeislinger/asl-rgb-depth-fingerspelling-spelling-it-out | | |
|
| Sign language MNIST | https://www.kaggle.com/datasets/datamunge/sign-language-mnist | | |
|
| Indian Sign Language | https://data.mendeley.com/datasets/rc349j45m5/1 doi: 10.17632/rc349j45m5.1 | | |
|
