Bhavaani Kavach: Autonomous Edge-AI Wearable Threat Detection & Offline Safety Engine
Inspiration
Two of my sisters are working professionals who frequently commute alone late in the evening. Worrying about their safety made me realize a critical flaw in current emergency solutions: during a sudden crisis, expecting a victim to unlock a phone, open an app, or press a panic button is unrealistic due to adrenaline freeze. Furthermore, standard safety apps fail completely when cellular data drops in basements or remote dead-zones. I created Bhavaani Kavach—named after the sacred Hindu Goddess Bhavaani, meaning "The Shield of Goddess Bhavaani"—to serve as an autonomous, offline-resilient guardian that protects users when they cannot manually react.
What it does
Bhavaani Kavach is a hands-free, zero-latency 360° threat detection wearable locket powered by embedded computer vision. It continuously monitors the wearer's immediate surroundings to detect sudden rear or frontal threats autonomously, eliminating human reaction delay. By pairing with a smartphone over Bluetooth Low Energy (BLE), it offloads heavy network and location tasks while keeping the wearable ultra-compact and energy-efficient.
When an anomaly is flagged, the system captures snapshot evidence and dispatches real-time location payloads. If cellular data (4G/5G) is unavailable, it automatically downgrades to direct satellite GPS tracking and SMS dispatch to guarantee emergency alerts reach contacts without internet.
Note on Dual-Use Utility: The core firmware can also be configured into a covert surveillance profile for female police officers and undercover operatives during sting operations, enabling silent local recording and discrete BLE broadcasting.
How we built it
Hardware Core: Built around an ESP32-CAM (OV2640) board programmed via an FTDI module for low-power vision processing, integrated with an I2S MEMS Microphone (INMP441) for acoustic validation.
Edge Intelligence Engine: Written in C++ for embedded microcontrollers and prototyped in Python using OpenCV. It utilizes real-time frame-difference and spatial anomaly algorithms to flag rapid physical motion locally without cloud processing.
Multi-Modal Verification Engine: Incorporates acoustic stress analysis via the I2S microphone to cross-reference motion spikes with vocal distress signals, preventing false positives during casual interactions.
Hybrid Connectivity Pipeline: Leverages low-latency BLE to pass trigger events to a mobile master engine.
Multi-Tiered Fallback Matrix: Features a dynamic network router:
4G/5G Cloud Upload->Direct GSM/SMS Dispatch->On-Board Encrypted SD Card Buffer.
Public Safety Integration: Formatted emergency telemetry payloads to align directly with national ERSS-112 (Emergency Response Support System) dispatch frameworks and AIS-140 telematics standards.
Manufacturing Economics & BOM To ensure the solution remains accessible to everyday commuting women across all economic backgrounds, we decoupled high-cost components by utilizing the smartphone's modem and GPS.
Estimated Unit BOM: ~₹890 ($10.70 USD)
ESP32-CAM Board (₹350) + OV2640 Lens (₹180) + INMP441 Mic (₹120) + 150mAh LiPo Battery (₹90) + Enclosure Casing (₹150)
Challenges we ran into
Resource Constraints: Optimizing OpenCV computer vision logic to run on microcontrollers with limited RAM without frame drops or high latency.
Power Efficiency vs. Continuous Monitoring: Balancing active camera sensing with battery life, solved by offloading cellular, GPS, and heavy processing to the paired smartphone via low-power BLE.
Zero-Network Reliability: Designing a fault-tolerant communication loop that seamlessly transitions from heavy image data uploads to lightweight satellite GPS + SMS payloads when coverage drops to zero.
Accomplishments that we're proud of
Zero-Human Latency: Developed a fully autonomous edge detection pipeline that removes manual intervention during panic scenarios.
True Offline Resilience: Engineered a multi-tiered fallback architecture operating reliably in dead-zones where standard safety apps fail.
Validated Software Baseline: Successfully verified the edge-AI anomaly engine logic through a functional Python simulation environment.
What we learned
Edge-AI Optimization: Techniques for running efficient real-time computer vision algorithms on resource-constrained embedded hardware.
Embedded Co-Design: How to structure a hybrid wearable-smartphone architecture using BLE to maximize battery efficiency and minimize physical form factor.
Fault-Tolerant Systems: Structuring fail-safe emergency communication pipelines for high-stakes, real-world deployment scenarios.
What's next for Bhavaani Kavach
Hardware Flashing: Flash the finalized C++ firmware directly onto physical ESP32-CAM hardware kits.
Modular Sensing: Integrate secondary rear-guard motion sensors for full 360-degree environmental awareness.
Form Factor Miniaturization: Design a custom 3D-printed enclosure packaging the board, battery, and haptic engine into a sleek, everyday wearable locket.
NOTEWORTHY:-
It can also be classically used for the Sting Operations, for the Police-women, as well, changing some part of the initial programming of the cam...!!
Log in or sign up for Devpost to join the conversation.