MUMBAI, India, Oct. 12 -- Intellectual Property India has published a patent application (202611103731 A) filed by Sohanlal Commodity Management Limited on August 27, 2026, for System And Method For Guided Mobile Capture, Multi-Frame Consensus Counting And Multi Source Based Verification Of Agricultural Commodity Stacks In Warehouses Using Standard Mobile Devices.
Inventor includes Sandeep Sabharwal.
The application for the patent was published on October 09, 2026, under issue no. 41/2026.
Abstract: A computer-implemented warehouse inventory-counting system and method are disclosed. A mobile device is guided to capture a physical tag (Stack Card) associated with a warehouse stack and subsequently record a video traversal of the stack. Captured video frames are evaluated for quality parameters including blur, brightness, stability, orientation and stack coverage, and a limited number of candidate frames are selected for processing by a computer-vision counting engine. Frame-level counts are aggregated using a median-based, outlier-rejecting consensus to generate a consolidated count, and a canonical evidence frame is selected. A tag-reading engine independently extracts a stack identifier and a claimed numerical count entered on a physical tag, and a deviation engine compares the consolidated count against the independently derived claimed numerical count and, optionally, a book-stock value, without automatically replacing the consolidated count with either. The comparison is used to generate an approval, review, recapture or physical-recount instruction, and a complete, timestamped, auditable digital record of the session, including the captured video, is stored for traceability and dispute resolution. In one embodiment, the system operates using only a standard mobile device, without necessitating a depth sensor, stereo camera, LiDAR sensor, fixed camera infrastructure or robotic capture platform. Warehouse- or commodity-specific model drift is detected using production indicators, and a candidate model is retrained using selected validated samples and historical baseline samples. The candidate model is evaluated, shadow-deployed and promoted or rejected based on accuracy, regression and stability criteria. A model repository records performance metrics for successive model versions and compares them to determine and record a measurable, auditable improvement in processing efficiency or accuracy over successive controlled retraining cycles, without degrading performance on previously encountered warehouse and commodity conditions.
Disclaimer: Curated by HT Syndication.
