# SignConnect: A Smart Glove System for Bidirectional Arabic Sign Language Communication > **Authors:** Amr Bedir · Karim El-Driny · Karoline Rushdy · Christine Amir · Rahma Hatem · Karim Gomaa · Sandy Tamer · Karim Agoor · Madonna Ibrahim · Maged Ashraf > > **Faculty of Computers and Information, Mansoura University, Egypt — 2026** > > **Supervised by:** Dr. Heba Kandil · Eng. Saher Walid --- ## Abstract Communication barriers between the Arabic-speaking Deaf and Hard-of-Hearing (HoH) community and the hearing world represent a persistent social challenge, with fewer than one certified Arabic Sign Language (ArSL) interpreter available per 10,000 Deaf individuals across Arab countries. Existing software-based assistive solutions rely on camera and computer vision approaches that fail in low-light conditions, compromise user privacy, and demand high computational resources. This paper presents **SignConnect** — a complete, bidirectional Arabic Sign Language communication system based on a custom-built Smart Glove. The glove integrates five resistive flex sensors and a BMI160 six-axis IMU (11 sensor axes total) driven by an ESP32 microcontroller transmitting via BLE 4.2+ at 33.3 Hz. Gesture recognition is implemented using an **Euclidean Distance Template Matching** algorithm operating on 100-frame, 3-second normalized sensor windows. The companion Flutter mobile application, published on Google Play, implements both sign-to-speech and speech-to-sign translation via a custom Blender 3D avatar. A community Data Collection Portal established the **first publicly available Arabic Sign Language video dataset** (34 words, 57 reviewed recordings). System evaluation confirmed BLE stability at 5 m+ range, end-to-end translation latency under 2 seconds, and a 4–6 hour battery life. Clinical validation through a 59-response healthcare professional survey confirmed 100% encounter rate with Deaf patients and >80% endorsement for a real-time translation application. **Keywords:** Arabic Sign Language · Smart Glove · Gesture Recognition · Euclidean Distance · BLE · Flutter · Bidirectional Communication · Deaf Assistive Technology · ArSL Dataset --- ## Table of Contents 1. [Introduction](#i-introduction) 2. [Related Work](#ii-related-work) 3. [Methodology](#iii-methodology) 4. [Experimental Results](#iv-experimental-results) 5. [Conclusion](#v-conclusion) 6. [Future Work](#vi-future-work) 7. [References](#references) --- ## I. Introduction Over 430 million people worldwide require hearing rehabilitation services, with more than 11 million in the Arab world and approximately 1.2 million in Egypt alone [1]. Arabic Sign Language (ArSL) is the natural language of this community, yet it remains inaccessible to the vast majority of the hearing population. The shortage of certified interpreters — estimated at a ratio of **1:10,000** — forces Deaf individuals to rely on inadequate workarounds such as paper-and-pen, informal family interpretation, or gesture-based improvisation in critical settings including healthcare consultations, educational institutions, and public service offices [2]. The healthcare impact of this barrier is significant. A survey of **59 medical professionals across 15 specialties** conducted as part of this project revealed that 100% of respondents had encountered Deaf patients, with consultation times averaging twice the standard duration. Over 80% of respondents rated a real-time sign language translation application as *"extremely helpful"* in their clinical practice — validating the urgency of the problem addressed by this work. Existing technological approaches to sign language recognition predominantly rely on computer vision: RGB cameras combined with hand detection, pose estimation (e.g., MediaPipe), or deep learning classifiers [3]. While these methods have achieved strong results on benchmark datasets, they suffer from fundamental limitations that restrict daily-use deployment: - Sensitivity to lighting conditions, background complexity, and camera angle - High battery and computational consumption - Continuous video recording that violates user privacy — particularly in clinical and educational contexts [4][5] Sensor-based wearable glove systems offer a fundamentally different paradigm, capturing hand gesture data directly from physical sensors without environmental dependencies. Prior work in this domain — including SignAloud [6], Mohandes et al. [7], and the BrightSign Glove [8] — has demonstrated the viability of the approach but left **three critical gaps** unaddressed for the Arabic-speaking Deaf community: 1. No Arabic Sign Language gesture recognition 2. No bidirectional communication (sign-to-speech AND speech-to-sign in a single system) 3. No publicly available Arabic Sign Language dataset for the research community This paper presents **SignConnect**, a complete bidirectional Arabic Sign Language communication system that addresses all three gaps simultaneously. The primary contributions of this work are: - **A custom 11-axis Smart Glove** (5 flex sensors + BMI160 6-axis IMU + ESP32 + BLE 4.2+) with per-user normalization calibration and standalone OLED/speaker output - **A production bidirectional ArSL mobile application** (Google Play: `com.signconnect.glove`) with Euclidean Distance Template Matching gesture recognition, 3D avatar speech-to-sign, and multi-provider authentication - **The first publicly available Arabic Sign Language video dataset**: 57 reviewed recordings covering 34 unique ArSL words, collected via a community volunteer portal - **A scalable gesture recognition pipeline** that supports vocabulary expansion without any model retraining — new signs added by recording sessions to a MySQL database --- ## II. Related Work ### Table I: Comparison of Related Systems | System | Approach | Language | Direction | Mobile | Key Limitation | |--------|----------|----------|-----------|--------|----------------| | Al-Qurishi et al. [4] | Camera / CNN | Arabic SL | Sign → Text | No | Limited vocab; not deployed; no dataset | | Aly & Aly [5] | Camera / SVM+CNN | Egyptian ArSL | Sign → Text | No | Static signs only; no real-time; no deployment | | SignAloud [6] | Sensor Glove (flex+accel) | ASL (English) | Sign → Text | No | English only; no avatar; no public dataset | | Mohandes et al. [7] | Sensor Glove (accel only) | Arabic SL | Sign → Text | No | No flex sensors; no bidirectional mode | | BrightSign Glove [8] | Sensor Glove (flex+IMU) | Multi-language | Sign → Text/Speech | Companion | No ArSL training; no avatar; no public app | | Abougarair & Arebi [9] | Sensor Glove (flex+accel) | ASL (English) | Sign → Text/Speech | Mobile | 6 signs; English only; no dataset | | **SignConnect (Proposed)** | **Sensor Glove (flex+BMI160) Euclidean Distance** | **Arabic SL** | **Bidirectional (Sign↔Speech)** | **Google Play** | **First public ArSL dataset; production app; open-source** | ### A. Camera-Based Arabic Sign Language Systems Al-Qurishi et al. [4] proposed a CNN-based ArSL recognition system achieving promising accuracy on a small vocabulary of static signs but limited to the sign-to-text direction, with no mobile deployment and no publicly released dataset. Aly and Aly [5] explored Egyptian ArSL using SVM and CNN classifiers but similarly addressed only static signs and did not provide real-time operation. Both systems retained the fundamental camera-based limitations described in Section I. ### B. Sensor-Based Sign Language Gloves SignAloud [6] from the University of Washington demonstrated real-time American Sign Language-to-speech translation using a glove with flex sensors and an accelerometer but was restricted to English (ASL), provided no avatar output, and released no dataset. Mohandes et al. [7] developed an Arabic sign language glove using accelerometers only — without flex sensors — limiting its ability to discriminate hand shapes from orientation. Abougarair and Arebi [9] demonstrated a complete sensor-glove-to-mobile pipeline with an initial 6-sign vocabulary but targeted ASL and provided no Arabic training data. ### C. BrightSign Glove The BrightSign Glove [8] (Ayoub, UK Patent GB2590502B, Goldsmiths University of London) represents the closest prior system to SignConnect in hardware philosophy: a self-contained sensor glove with flex sensors, IMU, on-device display, and multi-language output. The SignConnect team established **direct academic collaboration** with Eng. Hadeel Ayoub, who shared the technical patent documentation that informed the SignConnect hardware design. However, BrightSign does not train on ArSL gesture recognition, provides no speech-to-sign avatar output, and is not available as a free public mobile application. ### D. Gap Analysis No prior system has simultaneously provided: 1. Sensor-based (camera-free) ArSL gesture recognition 2. Bidirectional communication in a single application 3. A production mobile release 4. A publicly available Arabic Sign Language dataset **SignConnect addresses all four gaps concurrently.** --- ## III. Methodology ### Table II: SignConnect System Technical Specifications | Component | Specification | |-----------|--------------| | Microcontroller | ESP32 dual-core 240 MHz with integrated BLE 4.2+ | | Flex Sensors | 5× resistive flex sensors (Thumb, Index, Middle, Ring, Pinky) — analog ADC reading | | IMU | BMI160 (Bosch Sensortec) — 3-axis accelerometer + 3-axis gyroscope via I2C | | Total Sensor Axes | **11 axes** (5 flex + 3 accelerometer + 3 gyroscope) | | Sampling Rate | 33.3 Hz (30 ms BLE notification interval) | | Gesture Window | 3.0 seconds = **100 frames** per gesture | | Communication | Bluetooth Low Energy 4.2+ — device name: `SignConnect Glove` | | On-Glove Display | SH1106 1.3″ 128×64 OLED — 4 UI states (Welcome, Searching, Connected, Status) | | On-Glove Speaker | Miniature 5V speaker for standalone audio output | | Power | 3.7V Li-Po rechargeable battery — estimated 4–6 hours operation | | Mobile App | Flutter (Dart) — Android production release on Google Play (`com.signconnect.glove`) | | Backend | PHP 8.2 REST API on Hostinger VPS — MySQL 8.x (utf8mb4) | | Speech Recognition | Google Cloud Speech-to-Text API — Arabic (MSA + Egyptian dialect) | | 3D Avatar | Custom Blender 3.6 LTS character — 34 ArSL animations — WEBP format | | Authentication | Email/password (15-min verification) + Google OAuth 2.0 + Facebook SSO | --- ### A. Smart Glove Hardware The Smart Glove captures Arabic sign language gestures through an **11-axis sensor array** embedded in a wearable glove: - **Five resistive flex sensors** measure the curvature of each finger (Thumb, Index, Middle, Ring, Pinky) via analog ADC readings - **BMI160 six-axis MEMS IMU** (Bosch Sensortec) provides three-axis accelerometer and three-axis gyroscope data over I2C, capturing hand orientation and spatial motion vectors essential for distinguishing signs that share identical hand shapes but differ in movement direction or wrist rotation An **ESP32 microcontroller** reads all 11 sensor axes every 30 ms (33.3 Hz) and transmits formatted data packets via integrated BLE 4.2+ with a 30 ms notification interval. The glove advertises itself by device name (`SignConnect Glove`), enabling automatic discovery by the Flutter application without requiring manual device selection. Auxiliary components include: - **SH1106 1.3″ OLED display** (four UI states: Welcome, Searching, Connected, Status) - **Miniature speaker** — enabling standalone operation without the mobile application --- ### B. Per-User Calibration and Normalization Sensor output varies significantly across users with different hand sizes, finger lengths, and baseline muscle tension. SignConnect implements a **per-user calibration procedure** before first use: 1. User performs fully open hand posture (X_min) — system captures 30 frames → arithmetic mean per axis 2. User performs fully closed fist posture (X_max) — system captures 30 frames → arithmetic mean per axis 3. All subsequent sensor readings normalized using: $$X_{norm} = \frac{X_{raw} - X_{min}}{X_{max} - X_{min}}$$ Calibration baselines are stored per user in the MySQL database and retrieved at each session login, enabling **multi-user operation on the same physical device**. Normalization is applied to all 11 sensor axes individually. --- ### C. Gesture Recognition: Euclidean Distance Template Matching Following the 3-second gesture capture window (100 frames × 11 axes), the Flutter application: 1. Computes the **arithmetic mean** of each sensor axis across all 100 frames 2. Produces an **11-dimensional averaged feature vector** 3. Transmits the vector via HTTP POST to `predict.php` on the backend The prediction algorithm uses **Euclidean Distance Template Matching** against pre-computed master templates. Each master template represents a single ArSL gesture class, computed by: ```sql SELECT gesture_name_ar, AVG(thumb) as t, AVG(index_f) as i, AVG(middle) as m, AVG(ring) as r, AVG(pinky) as p, AVG(ax) as ax, AVG(ay) as ay, AVG(az) as az, AVG(gx) as gx, AVG(gy) as gy, AVG(gz) as gz FROM gesture_sessions s JOIN gesture_frames f ON s.id = f.session_id GROUP BY gesture_name_ar ``` For each incoming query vector **q = [t, i, m, r, p, ax, ay, az, gx, gy, gz]**, the Euclidean distance to each master template **T_k** is computed as: $$d(q, T_k) = \sqrt{\sum_{j=1}^{11}(q_j - T_{k,j})^2}$$ The gesture class with **minimum distance** is returned as the prediction, with confidence score: ``` confidence = max(0, 100 − (distance / 10)) % ``` **Key advantage:** New signs are added simply by recording new sessions to the database — **no model retraining, no redeployment, no code changes required.** The system improves automatically as more sessions accumulate. #### PHP Implementation (predict.php) ```php $best_match = 'Unknown'; $min_distance = INF; foreach ($master_templates as $template) { $distance = sqrt( pow($input['t'] - $template['t'], 2) + pow($input['i'] - $template['i'], 2) + pow($input['m'] - $template['m'], 2) + pow($input['r'] - $template['r'], 2) + pow($input['p'] - $template['p'], 2) + pow($input['ax'] - $template['ax'], 2) + pow($input['ay'] - $template['ay'], 2) + pow($input['az'] - $template['az'], 2) + pow($input['gx'] - $template['gx'], 2) + pow($input['gy'] - $template['gy'], 2) + pow($input['gz'] - $template['gz'], 2) ); if ($distance < $min_distance) { $min_distance = $distance; $best_match = $template['gesture_name_ar']; } } echo json_encode([ 'status' => 'success', 'prediction' => $best_match, 'distance_score' => round($min_distance, 2), 'confidence' => round(max(0, 100 - ($min_distance / 10)), 2) . '%' ]); ``` --- ### D. Mobile Application and Speech-to-Sign Direction The Flutter mobile application (MVVM + Provider architecture) implements both translation directions: **Sign → Speech:** ``` BLE Notification → 100-frame buffer → Normalize → predict.php → Arabic word → TTS ``` **Speech → Sign:** ``` Microphone → Google Cloud Speech-to-Text (Arabic) → Text → Avatar animation key → WEBP animation ``` The **3D avatar** (two custom characters: *Salem* and *Nora*, modeled in Blender 3.6 LTS with 62–64 bone armatures) is exported as transparent-background WEBP animated files. All animations are **pre-cached on app launch** for gapless zero-delay playback. An automatic **finger-spelling fallback** decomposes unrecognized words letter-by-letter using the 28-letter Arabic alphabet animations. #### ESP32 BLE Firmware (simplified) ```cpp #include <BMI160Gen.h> #include <BLEDevice.h> BLECharacteristic *pCharacteristic; void loop() { // Read flex sensors (analog ADC) int thumb = analogRead(THUMB_PIN); int index_ = analogRead(INDEX_PIN); int middle = analogRead(MIDDLE_PIN); int ring = analogRead(RING_PIN); int pinky = analogRead(PINKY_PIN); // Read BMI160 IMU via I2C int ax = BMI160.getAccelerationX(); int ay = BMI160.getAccelerationY(); int az = BMI160.getAccelerationZ(); int gx = BMI160.getRotationX(); int gy = BMI160.getRotationY(); int gz = BMI160.getRotationZ(); // Package and send as BLE 4.2+ notification String payload = String(thumb) + "," + String(index_) + "," + String(middle) + "," + String(ring) + "," + String(pinky) + "," + String(ax) + "," + String(ay) + "," + String(az) + "," + String(gx) + "," + String(gy) + "," + String(gz); pCharacteristic->setValue(payload.c_str()); pCharacteristic->notify(); // 30 ms BLE notification interval delay(30); // 33.3 Hz } ``` --- ### E. Data Collection Portal and ArSL Dataset A community volunteer portal ([signconnect.digital/Data-Collection-Portal](https://signconnect.digital/Data-Collection-Portal/)) was developed to address the near-complete absence of publicly available Arabic Sign Language datasets. **Volunteer workflow:** 1. Enter name, phone number, and target sign word 2. Select recording duration (3 / 5 / 7 / 10 / 15 seconds) 3. Record gesture via browser MediaRecorder API — no app installation required 4. Submit — explicit consent confirmation required before recording begins **Dataset status at time of writing:** - **57 reviewed video recordings** - **34 unique ArSL words** - **4+ volunteer contributors** - **Constitutes the first publicly accessible Arabic Sign Language video dataset** --- ### F. Iterative Prototyping Methodology Hardware development followed a **six-cycle Build-Test-Improve methodology**: | Cycle | Subsystem | Key Outcome | |-------|-----------|-------------| | 1 | Flex sensors only | Confirmed stable resistance-to-angle mapping; fixed ring finger mounting drift | | 2 | BMI160 IMU integration | Resolved I2C address conflict with OLED; confirmed clean 6-axis readings | | 3 | BLE communication | Determined 30 ms as optimal interval (empirically — lower caused buffer overflow) | | 4 | Flutter integration | Validated normalization pipeline and 100-frame circular buffer | | 5 | OLED display | Redesigned boot animation to meet <200 ms update latency target | | 6 | Full system integration | End-to-end testing under realistic wearing conditions | **Three hardware issues resolved during prototyping:** - Ring finger flex sensor position drift → repositioned along dorsal surface - BLE buffer overflow at <30 ms intervals → fixed at 30 ms - OLED boot animation exceeded 200 ms target → simplified state machine --- ## IV. Experimental Results ### Table III: System Evaluation Results | Evaluation Metric | Target | Achieved Result | |-------------------|--------|-----------------| | BLE notification interval | ~30 ms | ✅ 30 ms stable continuous stream | | BLE wireless range | ≥ 5 meters | ✅ Stable at 5 m+ (open area) | | OLED state update latency | < 200 ms | ✅ < 200 ms after boot animation redesign | | Mobile app frame rate | Stable 60 fps | ✅ 60 fps during live BLE data streaming | | End-to-end translation latency | < 2 seconds | ✅ < 2 s (gesture end → TTS audio output) | | Battery life | ≥ 4 hours continuous | ✅ 4–6 hours estimated | | Prototype iterations | — | 6 iterative cycles | | ArSL video dataset | First public ArSL dataset | ✅ 57 recordings, 34 unique ArSL words | | Vocabulary (production) | Expandable without retraining | ✅ 34 signs; new gestures via database only | | Healthcare survey endorsement | Validate clinical need | ✅ 100% encounter rate; >80% rated app "extremely helpful" | --- ### A. Hardware Performance The Smart Glove maintained a **stable BLE 4.2+ connection at distances exceeding 5 meters** in open-area testing, satisfying the minimum range requirement for practical indoor use in classrooms and clinical consultation rooms. The **30 ms BLE notification interval** was confirmed as the empirically optimal setting: - Intervals **below 30 ms** → BLE buffer overflow on the ESP32 - Intervals **above 30 ms** → perceptible lag in Flutter sensor visualization BMI160 IMU drift over 60 seconds of continuous operation was minimal and within acceptable bounds without mid-session recalibration. Battery life under continuous BLE streaming reached an estimated **4–6 hours** — sufficient for a full working day of intermittent use in educational or clinical settings. --- ### B. Software and Application Performance End-to-end translation latency — from gesture completion to TTS audio output — was consistently measured **below 2 seconds** under normal server load conditions, satisfying the real-time conversation requirement identified through the healthcare professional survey. The Flutter application maintained a stable **60 fps** rendering rate during live BLE data streaming and 3D avatar animation playback on standard mid-range Android devices. The Euclidean Distance Template Matching pipeline demonstrated its key practical advantage: **recognition capability improves automatically as more gesture sessions are recorded** to the database, without any code changes or model redeployment. The current production vocabulary of **34 signs** can be expanded indefinitely by any authenticated user through the Glove Data Stream Portal. --- ### C. Dataset and Community Contribution The SignConnect Data Collection Portal has collected **57 reviewed Arabic Sign Language video recordings** covering **34 unique ArSL words** from **4+ volunteer contributors** at the time of writing. This constitutes the **first publicly accessible ArSL video dataset**, directly addressing a research gap identified across all reviewed prior work. The portal remains live and actively collecting contributions at: ``` https://signconnect.digital/Data-Collection-Portal/ ``` --- ### D. Clinical and Social Validation A Healthcare Professionals Survey collected **59 validated responses** spanning **15 medical specialties** including General Medicine, ENT, Dentistry, Speech Therapy, Neurology, Cardiology, Radiology, and Pharmacy. **Key findings:** | Finding | Result | |---------|--------| | Respondents who encountered Deaf patients | **100%** | | Average consultation time increase with Deaf patients | **~2× standard duration** | | Rated translation app as "extremely helpful" or "helpful" | **> 80%** | | Prioritized accuracy over speed | **> 75%** | > The >75% accuracy preference directly validates the **Euclidean Distance Template Matching** approach (high precision template matching) over faster but less reliable real-time neural network classifiers. A Public Awareness Survey distributed via social media confirmed widespread public unawareness of ArSL and broad demand for accessible communication technology. --- ## V. Conclusion This paper presented **SignConnect** — a complete, bidirectional Arabic Sign Language communication system that addresses three critical gaps left unresolved by existing work: 1. **Sensor-based ArSL gesture recognition** — eliminating camera dependency entirely 2. **Bidirectional communication** in a single production deployment 3. **The first publicly available Arabic Sign Language video dataset** The custom Smart Glove (ESP32, 5× flex sensors, BMI160 IMU, BLE 4.2+, SH1106 OLED) with per-user normalization calibration provides reliable 11-axis sensor capture in any lighting condition and at distances up to 5 m+ with **under 2-second end-to-end translation latency**. The Euclidean Distance Template Matching pipeline enables **incremental vocabulary expansion without retraining**, currently supporting 34 ArSL signs. The companion Flutter application — published as a production release on Google Play — implements both sign-to-speech and speech-to-sign translation via a custom Blender 3D avatar with sentence-level processing and automatic finger-spelling fallback. Clinical validation through 59 healthcare professional responses confirmed the real-world urgency of the problem and strong clinical demand for the solution. **The system is fully deployed, open-source, and freely accessible** — serving its target community from day one of publication. --- ## VI. Future Work | Direction | Description | |-----------|-------------| | **Vocabulary Expansion** | Grow gesture dictionary from 34 to 500+ signs (daily, medical, educational, emergency vocabulary) via continued community data collection | | **Deep Learning Replacement** | Replace Euclidean Distance classifier with LSTM or Transformer-based temporal sequence model as dataset grows beyond hundreds of examples per class | | **On-Device Edge Inference** | Port recognition model to TensorFlow Lite for on-device inference on ESP32-S3 — enabling true offline, zero-latency translation | | **Hardware Miniaturization** | Develop next-generation fully integrated single-PCB wristband combining ESP32, IMU, BLE, OLED, and battery management | | **iOS Release & Dialect Support** | Deploy Flutter app on Apple App Store; extend gesture dictionary to support Jordanian, Saudi, and Gulf ArSL dialect variants | | **Clinical Integration** | B2G/B2B partnership with hospital management systems (HIS/EMR) in Egyptian and Arab healthcare facilities with pre-loaded medical vocabulary packs | --- ## References [1] World Health Organization, "World Report on Hearing," WHO Press, Geneva, 2021. [2] Egyptian Ministry of Social Solidarity, "Annual Report on Persons with Disabilities in Egypt," Cairo, 2022. [3] K. Bantupalli and Y. Xie, "American Sign Language Recognition Using Deep Learning and Computer Vision," in *Proc. IEEE Int. Conf. Big Data*, 2018, pp. 4896–4899. [4] M. Al-Qurishi et al., "DeepDistributed: A Deep Learning Model for Arabic Sign Language Recognition," *Future Generation Computer Systems*, vol. 114, pp. 1–9, 2020. [5] S. Aly and A. Aly, "Deeparsl: Egyptian Arabic Sign Language Recognition Using Deep Learning," *IEEE Access*, vol. 8, pp. 83199–83210, 2020. [6] T. Huang and J. Weitz, "SignAloud: Gloves That Transliterate Hand Gestures Into Text and Speech," Univ. of Washington, Lemelson-MIT Student Prize, 2016. [7] M. Mohandes, M. Deriche, and J. Liu, "Image-Based and Sensor-Based Approaches to Arabic Sign Language Recognition," *IEEE Trans. Human-Machine Systems*, vol. 44, no. 3, pp. 335–346, 2014. [8] H. D. Ayoub, "Sign Language Translation Glove Device," UK Patent GB2590502B, Goldsmiths, Univ. of London, 2021. [9] Y. J. Abougarair and M. K. Arebi, "Smart Glove for Sign Language Translation," *Int. Robotics and Automation Technology Journal*, vol. 8, no. 3, 2022. doi: 10.15406/iratj.2022.08.00253. [10] Bosch Sensortec, "BMI160 Data Sheet," Rev. 1.2, Bosch Sensortec GmbH, Reutlingen, 2015. [11] Espressif Systems, "ESP32 Technical Reference Manual v5.0," Shanghai, 2023. [12] SignConnect Project, "Data Collection Portal," https://signconnect.digital/Data-Collection-Portal/, 2026. --- <div align="center"> **© 2026 SignConnect Team — Faculty of Computers and Information, Mansoura University** *"Bridging the gap, one sign at a time."* [](https://play.google.com/store/apps/details?id=com.signconnect.glove) [](https://signconnect.digital/Data-Collection-Portal/) [](https://wiki.signconnect.digital) </div>