The Department of Navy (DON) needs an inexpensive lidar-like 3D imaging sensors that have high depth and lateral resolution, have a large field-of view for reliable object detection, respond in real time, and work at medium to long ranges in indoor and outdoor environments. This need can be met by combining stereovision with an adaptive laser range finding array that can selectively increase the resolution of the point clouds in desired regions of interest at far-off distances. This proposal presents a plan to develop a Target Detection and Adaptive Ranging (TDAR) sensor with advanced Machine Learning (ML) based target detection algorithms-in-the-loop. This approach effectively combines the advantages (FOV, fps, point-cloud data) of the vision-based and LIDAR-based target detections, while omitting its drawbacks (stereo disparity, computationally expensive algorithms, expensive hardware) to achieve high resolution point cloud data at near and far ranges. The innovation of the TDAR sensor is the adaptive ranging that only generates high-resolution point cloud data of relevant targets that are rapidly detected by the stereo vision. This advantage allows the TDAR sensor to have low cost, SWAP, and computation requirements, in addition to the increased resolution.
Benefit: The TDAR sensor is capable of accurately detecting a wide range of targets at near and far-off distances in near real-time. A single or multiple of these units can be tailored to suit any application, including but not limited to defense, transportation, manufacturing, agriculture, automation, and pharmaceutics. With no mechanical components and only generating point cloud data of relevant targets, the component and computational cost of these sensors is much lower, making them attractive to self-driving cars due to cost, small autonomous systems due to SWAP, and factory automation where precise resolution is needed for far-off range.
Keywords: Machine Learning, Machine Learning, Target Detection, stereovision, 2D Image Detection, Adaptive Laser Ranging
