MUMBAI, India, Sept. 28 -- Intellectual Property India has published a patent application (202641112115 A) filed by Vallurupalli Nageswara Rao Vignana Jyothi Institute Of Engineering And Technology on September 18, 2026, for Real-Time Deep Learning Framework For Road Scene Analysis And Obstacle Identification In Autonomous Driving Applications.
Inventor includes Mr. E. Gurumohan Rao.
The application for the patent was published on September 25, 2026, under issue no. 39/2026.
Abstract: ABSTRACT [0025] The invention provides a semantic segmentation system tailored for road scenes in autonomous driving. The system employs a hybrid architecture that integrates U-Net’s encoder-decoder structure with a ResNet backbone as the feature extractor. This combination allows precise pixel-wise classification while extracting deep hierarchical features without suffering from vanishing gradients. Input road images or video frames undergo preprocessing including resizing, normalization, and data augmentation. The trained model produces color-coded segmentation masks that identify roads, vehicles, pedestrians, traffic signs, and other elements. Training uses diverse datasets covering urban, suburban, and highway scenes under varying lighting and weather conditions. Optimization techniques such as mixed-precision training, batch normalization, and conversion to efficient runtime formats enable near real-time inference suitable for vehicle perception pipelines. The resulting segmentation maps support downstream tasks including path planning and obstacle avoidance, thereby improving safety, reliability, and responsiveness of autonomous vehicles in dynamic road environments.
Disclaimer: Curated by HT Syndication.
