Deep Learning-based Vehicle Tracking with Multiple Sensors

Deep Learning-based Vehicle Tracking with Multiple Sensors 3/19/2018 The DOT is seeking an integrated unit that is capable of identifying and tracking individual vehicles with high accuracy using multiple vehicle detection inputs, including video images, radar-based detection, and wireless signal inputs (including DSRC, cellular, Wi-Fi, and Bluetooth signals), with a deep learning-based engine to fuse the collected data. To address this DOT need, Physical Optics Corporation (POC) proposes to develop a new Deep Learning-based Vehicle Tracking with Multiple Sensors (DETRAM) system based on a deep learning engine. DETRAM uses multiple deep neural networks to extract intelligence from a diverse set of sensors and fuses information from this variety of signal sources at both feature-level and decision-level of the processing pipeline to identify and track individual vehicles under all environmental conditions with a precision that is unmatched by state-of-the-art systems that use a subset of these signal sources. During Phase I, POC will demonstrate DETRAM’s feasibility with a proof-of-concept design that will include a concept of operations, systems requirement, systems design, and data management platform. In Phase II, POC will develop a prototype to perform field operational tests of DETRAM in coordination with a state or local highway agency.