MUMBAI, India, July 30 -- Intellectual Property India has published a patent application (202621067531 A) filed by Dr. Kaushalya Thopate; Ram Ashok Kulkarni; Soham Ladkat; Krishnapriya Sharma; Kumaresh Mondal; Anushka Kumbhar; Madhavi Mohite; Sheetal Sobale; Dinesh Washimkar; Vishwakarma Institute Of Technology, Pune; Vijay Mane; Pooja Gavhane; and Vaishali Patil on May 29, 2026, for A Non-Invasive Multi-Layer Human Presence Confirmation And Helmet Detection System For Power-Efficient Two-Wheeler Ignition Control.
Inventors include Dr. Kaushalya Thopate; Ram Ashok Kulkarni; Soham Ladkat; Krishnapriya Sharma; Kumaresh Mondal; Anushka Kumbhar; Madhavi Mohite; Sheetal Sobale; Dinesh Washimkar; Vijay Mane; Pooja Gavhane; and Vaishali Patil.
The application for the patent was published on July 24, 2026, under issue no. 30/2026.
Abstract: The present invention discloses a non-invasive multi-layer human presence confirmation and helmet detection system for power-efficient two-wheeler ignition control. The invention addresses the critical road safety problem of helmet non-compliance among two-wheeler riders by implementing a four-layer cascaded detection architecture that prevents vehicle ignition unless both rider presence and helmet compliance are simultaneously confirmed. The system operates through four sequentially gated detection layers. The first layer employs an ultrasonic distance sensor operating continuously at minimal quiescent power as a system wake-up trigger, detecting presence of any object within a predefined proximity threshold of the vehicle seat. The second layer employs a Force Sensitive Resistor sensor connected to a microcontroller analog input via a voltage divider circuit, independently confirming actual human body weight and eliminating false positive triggers caused by inanimate objects including luggage bags, backpacks, and other non-human items placed on the seat. Upon simultaneous positive confirmation by both hardware layers, the third layer activates the image capture device which was maintained in a completely inactive and unpowered state during idle operation and employs a pretrained YOLOv8 nano deep learning model to verify actual human presence within the camera frame. The fourth layer subsequently employs a separately trained custom YOLOv8 deep learning model, trained for a minimum of one hundred epochs on a domain-specific helmet dataset, to perform helmet compliance classification across a minimum of ten sequential camera frames with an early exit mechanism that terminates frame analysis immediately upon detection of helmet compliance in even a single frame within the ten-frame sampling window. An Arduino Uno microcontroller manages all hardware sensor reading and implements bidirectional serial communication with a Python-based external processing unit via a USB serial interface using a state-based command protocol. The ignition control actuator, represented by a Light Emitting Diode in the prototype and replaceable by an electromechanical relay in production deployment, is activated exclusively upon simultaneous positive confirmation of all four detection layers. Failure of any single detection layer immediately prevents ignition actuator activation and automatically resets the system to initial idle scanning state. Additionally, the system implements continuous post-ignition monitoring via the Force Sensitive Resistor sensor to detect rider departure from the seat, upon which the ignition actuator is immediately deactivated, the image capture device is powered down, and the system automatically resets to the initial idle scanning state, thereby preventing unauthorized vehicle operation by any subsequent non-helmet-wearing rider. The key novelty of the present invention lies in its cascaded power-efficient architecture wherein each successive detection layer consumes progressively higher power than its predecessor, with the highest power consuming components comprising the image capture device and neural network inference pipeline remaining completely inactive during the statistically most frequent idle vehicle state. This architecture simultaneously achieves significant power conservation, elimination of false positive detections through multi-modal sensor fusion, and superior helmet detection accuracy through the use of two independently trained and independently optimized deep learning models operating in sequential confirmation-gated execution. The invention is entirely non-invasive, requiring no physical modification to the rider's helmet or the two-wheeler vehicle, and is universally deployable across all two-wheeler types and helmet variants. The system is operable on a standard personal computer or laptop without requiring dedicated embedded hardware for deep learning inference.
Disclaimer: Curated by HT Syndication.
