Silent Alert: Advancing Women’s Security through Smart Sign Recognition and AI

Aim:

        To develop a real-time video-level-sign classification system that identifies rescue and emergency hand signs using BiLSTM, enabling automated alert messages to guardians via Twilio SMS.

Abstract:

         Women’s safety has become a global concern, and innovative technological solutions are being developed to address this challenge. This project proposes a real-time video-level-sign hand gesture recognition system to classify “rescue” and “emergency” signs. Inspired by a TikTok rescue sign used in a real-life incident, the system captures and processes video frames to extract hand landmarks using MediaPipe. A BiLSTM model is trained on landmark sequences to classify gestures as either “rescue” or “emergency”.

          The live system triggers an alert via Twilio SMS to a registered guardian when a “rescue” gesture is detected. Extensive experimentation shows that the proposed system outperforms existing solutions in terms of accuracy, robustness, and real-time responsiveness.

Introduction:

       The increase in crimes against women and children has highlighted the urgent need for effective safety measures. Traditional methods, such as panic buttons and wearable devices, require manual activation, which may not always be possible during emergencies. This project aims to address this issue by developing an AI-based system that automatically detects and classifies hand gestures in live video feeds.

          The proposed system identifies two gesture classes: “rescue” and “emergency.” By leveraging MediaPipe for landmark extraction and BiLSTM for temporal sequence modeling, the system achieves superior accuracy compared to existing methods. Once a “rescue” gesture is detected, an alert is sent to a parent or guide via Twilio SMS, providing location information for prompt assistance.

Existing System:

          Existing systems for gesture recognition rely heavily on static frame-based approaches. These systems often fail to capture the temporal information inherent in video data, leading to suboptimal classification accuracy. Previous models have used basic classifiers such as Random Forest, Decision Trees, and Support Vector Machines (SVM), which lack the ability to effectively model temporal dependencies in hand gestures.

Disadvantages:

• Limited Temporal Awareness: Static frame-based models do not utilize the sequential nature of video frames, leading to poor performance on dynamic gestures.
• Frame-Level Detection Only: Existing systems can only predict gestures frame-by-frame, lacking the capability to understand multi-frame context.
• Manual Activation Required: Existing safety alert systems rely on manual activation, which may not be feasible during emergencies.
• Lack of Real-Time Processing: Many existing models operate offline, delaying critical alerts.
• No Map Access: Existing systems do not provide access to GPS-based map information to assist in locating the user in emergencies.

Proposed System:

       The proposed system is a video-level gesture recognition model aimed at improving the safety of women and children. It utilizes MediaPipe for precise hand landmark extraction and a BiLSTM model to classify gestures into two categories: “rescue” and “emergency.” Unlike existing systems that predict frame-by-frame, this system processes the entire video sequence, allowing for more accurate recognition of dynamic gestures. The “rescue” gesture, inspired by a real-life TikTok-based rescue sign, plays a critical role in the system. When this gesture is detected, a live video feed is monitored continuously, and an alert message with GPS location access is sent to the guardian’s mobile device via Twilio SMS, ensuring timely and efficient assistance.