aiMotive Dataset: A Multimodal Dataset for Robust Autonomous Driving with Long-Range Perception (arxiv 2022/11)

Summary

https://arxiv.org/pdf/2211.09445.pdf
https://github.com/aimotive/aimotive_dataset/tree/f71828446692587318ebccbd3cdad5b4335eb9f3 datasetに関して
https://github.com/aimotive/aimotive-dataset-loader api
https://github.com/aimotive/mm_training
Camera-LiDAR-Radar dataset の提案
- 200mまでannotationされているので遠距離detectionに用いることができる
- multi-modalなbaselineも公開されているが、Camera以外少ししょぼい印象

64-beam top LiDAR
- 1x LiDAR Spinning, 64 beams, 10Hz capture frequency, 360◦ horizontal FOV, +15◦ to -25◦, vertical FOV, ≤ 200m range, ± 2 cm accuracy, up to 1.15 Million Points per Second
- Hesai Pandar64
4 cameras
- 2 （前後）pinhole: 30 to 40 Hz capture frequency, 1/1.55” CMOS sensor of 2896 × 1876 resolution
  - IMX490
- 2 （左右）fisheye: 30 to 60 Hz capture frequency, 1/2.7” CMOS sensor of 1920 × 1080 resolution
2 long-range radars
- 77GHz Radar
  - 18 Hz capture frequency, -16° to +16° azimuth FOV, -7.5 to +7.5 elevation FOV, up to 175m distance, velocity accuracy of ±0.1 m/s
  - UMRR11
GNSS+INS
- up to 100 Hz measurement frequency, position accuracy of 100 mm (PPK), heading accuracy of 0.044◦, roll&pitch accuracy of 0.009◦
mapは入っていないっぽい