Datasets

ARIL dataset

The Acoustic Recognition based Invisible-target Localization (ARIL) dataset is procured to advance the study of a model dedicated to the detection and localization of Non-Line-of-Sight (NLoS) vehicle. The dataset is centered on identifying NLoS vehicles, and encompasses variations in velocity (5, 10, 15, 20 km/h), the direction of travel for NLoS vehicles (left, right), and different spatial arrangements (T-Junction configurations, with and without opposite-side wall).

ARIL dataset is amassed within a carefully constructed test bed designed to mirror the conditions encountered on actual roads. To ensure meticulous ground truth data collection and to encapsulate the entirety of the test bed within a singular view, a 12-megapixel camera equipped with a fisheye lens offering a 185° field of view was utilized as the Bird's Eye View (BEV) Camera. This camera was strategically mounted at an elevation of 7 m above the intersection’s center within each spatial layout. An SUV outfitted with a microphone array, and camera, served as the primary vehicle for data collection. Location of NLoS vehicle was acquired from Bird's Eye View (BEV) camera. The total acquired data size is 11.7 TB.

Each scenario is documented through a series of data captures and location of NLoS vehicle: one sound sequence, one BEV image sequence, and one .xlsx file. The audio data was captured at a fidelity of 48,000 Hz, while BEV image data was collected at a frequency of 10 Hz. [Code][Dataset]

• Mingu Jeon, Jae-Kyung Cho, Hee-Yeun Kim, Byeonggyu Park, Seung-Woo Seo, and Seong-Woo Kim, "Non-Line-of-Sight Vehicle Localization based on Sound," under 1st revision for IEEE transactions on Intelligent Transportation Systems