MUMBAI, India, June 30 -- Intellectual Property India has published a patent application (202641054399 A) filed by Dr S Prakash; Rithin A R; Pradeep M K; Praveen Kumar S; and Prayin Kumaar T G on April 29, 2026, for Got-Chu Similar Object Detection System.
Inventors include Dr. S. Prakash; Rithin A R; Pradeep M K; Praveen Kumar S; and Prayin Kumaar T G.
The application for the patent was published on June 26, 2026, under issue no. 26/2026.
Abstract: The GOT-CHU Similar Object Detection System is an advanced computer vision solution designed to identify and locate objects in an image based on their visual similarity to a given sample image, rather than relying on predefined object classes. This makes the system highly flexible and adaptable, as it can detect previously unseen objects without requiring retraining. The core idea is to enable dynamic object recognition by analyzing visual features such as shape, texture, and structural patterns, which are critical for similarity-based detection tasks. The system operates through a structured pipeline that includes image preprocessing, object segmentation, detection, feature extraction, and similarity matching. Initially, SAM Mobile is used to segment the image and isolate potential object regions, even in complex backgrounds. These segmented regions are then refined using RF-DETR, a transformerbased detection model that generates accurate bounding boxes and improves localization, especially in cases involving overlapping or densely packed objects. To determine similarity, the system extracts deep feature embeddings from both the sample image and detected object regions using deep learning techniques. These embeddings are compared using cosine similarity, which measures how closely the objects resemble the sample in a high-dimensional feature space. A predefined similarity threshold is applied to identify matching objects, ensuring both accuracy and robustness. Supporting libraries such as PyTorch, OpenCV, NumPy, and Scikit-learn are used to handle computation, image processing, and model execution efficiently. 29-Apr-2026/63621/202641054399/Form 2(Title Page) Overall, the proposed system effectively addresses key challenges such as scale variation, occlusion, and complex visual environments. It is capable of handling high-resolution images and delivers reliable detection results suitable for real-world applications like industrial inspection, inventory management, robotics, and automation. By combining segmentation, transformer-based detection, and embedding-level similarity matching, the system demonstrates a scalable and intelligent approach to modern object detection problems.
Disclaimer: Curated by HT Syndication.
