Sensors


Sensor-kit Overview Loading...


The Vehicle-Infrastructure platform is shown in Fig. 1. Our onboard sensor suite consists of a multi-camera setup (event camera, industrial camera, and infrared camera) equipped with three LiDARs, onboard IMU, two GNSS receivers, and GNSS-RTK/INS systems. The specific specifications of each sensor are presented in Table. 1. . An Intel NUC (i7-1260P, 32GB RAM) and an industrial computer (i7-10610U, 32GB RAM) are used to run sensor drivers, and record data into ROS bags on the Ubuntu system.

Image description
Fig. 1. An overview of the Vehicle-Infrastructure sensor setup

Vehicle-Infrastructure sensor platform

The details of full sensors setup can be found below:

PlatformSensor TypeDescription
Onboard Platform Hikvision MV-CS050-10GC PRO × 7
  • Resolution: 2220 × 1740 pixels @ 10 fps
  • H-FOV: 80.18°, V-FOV: 73.77°
Hesai XT-32
  • H-FOV: 360°, V-FOV: +15° ~ -16°
  • 32 channels, 10 Hz
  • Range: 125 m
Velodyne VLP-32E
  • H-FOV: 360°, V-FOV: -30.67° ~ +10.67°
  • 32 channels, 10 Hz
  • Range: 100 m
GPAL-Ares-R7861 Radar
  • H-FOV: 150°, V-FOV: 30°
  • Front-view radar
  • Frequency: 13 Hz
Xsens MTi-30 IMU
  • Sampling Rate: 400 Hz
  • Accelerometer in-run bias instability: 15 µg
  • Gyroscope in-run bias instability: 18°/h
U-Blox ZED-F9P
  • Automotive Grade GNSS Receiver
  • 1 Hz, GPS Time
NovAtel SPAN-CPT
  • Geodetic Grade GNSS/INS System
  • 1 Hz, GPS Time
  • Position Accuracy: 5 cm
Roadside Infrastructure Innovusion Jaguar LiDAR
  • 300 Lines
U-Blox ZED-F9P
  • Automotive Grade GNSS Receiver
  • 1 Hz, GPS Time
UWB Nooploop
  • Model: P-B
  • Frequency: 50 Hz

We use an Ethernet topology for data transmission and time synchronization, as shown in Fig. 2. All independent devices and the PTP server are synchronized using NMEA and PPS signals from a u-blox M8T GNSS receiver. PTP server synchronizes devices clocks by broadcasting PTPv2 messages. Ethernet cameras are triggered via GigE Vision with a timing accuracy of < 0.005s.

Image description
Fig. 2. System architecture

Visual sensors

Seven Hikvision MV-CS050-10GC PRO cameras with 4mm lenses capture 2200×1740 RGB images at 10 Hz. Two are forward-facing stereo cameras, while five form a 360° surround setup with 60° spacing and 20° overlap. The FOV coverage is shown in Fig. 3 Image synchronized trigger is achieved via the GigE Vision protocol.

Image description
Fig. 3. Cameras FOV Coverage

4D Radar

A front-mounted 4D radar GPAL Ares-R7861 measures object position, velocity, and distance in real time, producing 3D point clouds with velocity data. It offers a 150° horizontal FOV and operates at 13 Hz. The point cloud view can be found in Fig. 4

Image description
Fig. 4. Radar sample

Mechanical LiDAR

Two mechanical LiDARs are used: a Velodyne HDL-32E on the vehicle roof for horizontal surroundings and a slanted Hesai XT-32 at the front for detailed urban architecture. Two LiDAR point clouds are fused as Fig. 5. Both operate at 10 Hz.

Image description
Fig. 5. LiDAR sample

Roadside LiDAR

Innovusion Jaguar LiDAR at 300 Lines cover the roundabout areas

Image description
Fig. 6. Roadside LiDAR

Roadside UWB

We deployed 4 UWB anchors with RSI, which ensure the communication between vehicle and RSU at the roundabout area.

Image description
Fig. 7. Roadside UWB