Setup

1611138600052011110.png

The MPD data is collected by the built multi-modal awareness vehicle system. The data is recorded in real time by the industrial computer equipped with GPU. Six cameras and four LiDARs are set on top of the vehicle, and here is the sensors setup diagrams:

图片6.jpg

The vehicle speed is maintained at 40km / h during data collection. OpenMPD consists of consecutive frame fragments, and each fragment is 5 seconds. For continuous image frames within 5 seconds, the camera annotates every other frame to obtain 100 labeled images. In addition, the mechanical lidar annotates each point cloud continuously within 5 seconds. To help researchers quickly become familiar with OpenMPD, tutorials for using OpenMPD are provided.

camera

The OpenMPD dataset is designed for a complex situation in the real world. In this condition, we need the views for our computer to be real 360° as what a driver can see. Therefore, we update our method from what nuScene had, we add one more camera which faces the backside of the vehicle. It just likes what drivers can see from the rearview mirror.

Below is the parameters for all cameras:

Sensors/parameter

Number

Resolution

Focal length

FOV

Data propagation speed

Range

Front_middle

1

1024*968

25mm

23°

47.41Mbp/s

50-150m

Front_Left & Right

2

1024*968

6mm

100°

47.41Mbp/s

~50m

Behind

1

1024*968

8mm

65°

44.40Mbp/s

~100m

Side_Left & Right

2

1024*968

5mm

78°

16.41Mbp/s

~3m

LiDAR

In order to provide a 360° coverage of the whole environment with LiDAR. The LiDAR with 128 beams located at the front rooftop of the vehicle is used as the main scene acquisition device. Comparing to those LiDARs with fewer beams, it can collect information from a further distance, and can obtain more accurate data and a denser number of points. However, it has a larger blind spot as well. To fix this problem, we decided to use 2 LiDAR with 16 beams on the sides and 1 LiDAR with 40 beams on the back to cover the blind spot. Note that we set all LiDAR to the frequency 10Hz.

Below is the parameters for all LiDARs:

Sensors/parameters

Number

Beam

Vertical FOV

Vertical Resolution

Range

(offical

Range

(tested

Front

1

128

-25° +15°

0. 1°

200m

150m

Behind

1

40

-25° +15°

0.33°

200m

100m

Side_Left & Side_Right

2

16

-15° +15°

2°

100m

30m

Thanks to Datatang(Stock Code : 831428) for providing us with professional data annotation services. Datatang is the world’s leading AI data service provider. We have accumulated numerous training datasets and provide on-demand data collection and annotation services. Our Data++ platform could greatly reduce data processing costs by integrating automatic annotation tools. Datatang provides high-quality training data to 1,000+ companies worldwide and helps them improve the performance of AI models.

Tsinghua University

Haidian District, Beijing, 100084, P. R. China

Email:xyzhang@tsinghua.edu.cn

© Copyright 2016-2020 www.OpenMPD.com. All Rights Reserved