Recording Setup

The E-VLC dataset focuses on the scenes where the receiver (cameras) moves at a walking speed, while the scene (including LED markers) is static. To this end, we build a custom mobile recording system equipped with two high-resolution cameras, a tablet, and a trigger box, enabling us to achieve high-quality calibration and synchronization. Our custom-built system consists of Prophesee EVK-4, 1280 x 720 pix and Basler acA1300-200um, 1280 x 1024 pix. Both cameras have a slight baseline (i.e., stereo settings) and are recorded to a tablet (Surface 11 Pro, Ubuntu 22.02) via USB cables.

The frames are hardware-triggered and record videos at 40fps. The trigger box sends synchronization pulses at 120Hz to three different destinations: the two cameras and an external motion capture (OptiTrack). We design one of the trigger cables to be 10m so that it allows the recording device to move inside the motion capture system, which provides GT positions of the camera and LED markers at sub-millimeter accuracy at 100Hz. Notice that there are small offsets between the cameras and the markers of the motion capture.

As LED markers (Orange circles), we use bullet-shaped LEDs (OSPG5111A, Green, 0.1W) with a custom microcontroller that controls the LED modulation. One microcontroller controls five LEDs with different blink patterns, whose timings (temporal resolution) are synchronized. The data transmitted are a few bytes of numbers (IDs), and each LED repeats the assigned blink pattern, which is typically around 10ms. In total, we use 40 LED markers (i.e., 8 microcontrollers), all of which transmit different IDs. The markers are attached to different objects, such as boxes and tables, as well as an Aruco marker (printed on A1 size) to benchmark with frame-based localization methods. The precise LED positions are measured with the motion capture system before the camera recordings (Yellow circles).