MUMBAI, India, July 24 -- Intellectual Property India has published a patent application (202611059708 A) filed by Delhi Technological University on May 11, 2026, for System And Method For Cross-Modal Pedestrian Behavior Prediction With Dual-Task Learning.
Inventors include Sharma, Neha; Dhiman, Chhavi; and Indu, S..
The application for the patent was published on July 17, 2026, under issue no. 29/2026.
Abstract: The present disclosure relates to a dual-task system (100) and method (200) for pedestrian behavior prediction, comprising short-term intention estimation and long-term trajectory forecasting. The system (100) acquires multimodal inputs, including pedestrian ROI images, segmentation maps, and past trajectory coordinates, and encodes motion history into a trajectory embedding. Trajectory- conditioned transformer networks with Progressive Denoising Attention (PDA) extract cross-modal features, which are fused and refined through a Context-Aware Feature Fusion Module (CAFFM) (118) to generate context-aware embeddings. A Variational Autoencoder (VAE), conditioned on these embeddings, predicts future pedestrian trajectories over multiple time steps. By integrating visual, semantic, and temporal information, the system (100) achieves robust, accurate, and interpretable predictions, enhancing generalization to diverse environments and enabling proactive decision-making in autonomous navigation, traffic safety, and surveillance applications.
Disclaimer: Curated by HT Syndication.
