Infrastructure and Data 2018

ABOUT THE PROJECT
 

At a glance

This document describes the scope of work for the BDD Infrastructure project. It gives a summary of the core activities on data sets and experimental vehicles to support BDD research. The planned activities for the infrastructure project in 2019 include:

• Continue image annotation work on BDD-100K, and to prepare for future data release to sponsors and the public; The task is covered by a separate data project, led by Fisher Yu and supervised by Prof. Trevor Darrell; Please see updates on the BDD 100k page, bdd-data.berkeley.edu
• Acquire and process vehicle-based driving data collected and to be provided by Holomatic, a BDD sponsor; data elements include lidar, radar, camera, and CAN-bus information, supplemented with image annotations. The infrastructure team will coordinate the participation and involvement of multiple research teams within BDD who are potential users of this vehicle-based data set from Holomatic.
• Maintain experimental vehicle hardware and software platforms;
• Continue support of vehicle usage by BDD research groups
• Continue analysis and modeling of pedestrian-traffic data set collected on Hearst Street near campus

Infrastructure Core Activities for 2019

(1) Activities for Experimental Vehicle Platforms MKZ and MKS (I-Ming Chen)

The BDD infrastructure team is responsible for the management and maintenance of two experimental vehicles: the 2017 Lincoln MKZ and 2013 Lincoln MKS. The regular tasks include:

• Basic maintenance duties: fueling, tire pressure check, periodical inspection, battery and power system maintenance;
• On-board device or sub-system firmware update;
• Tutorials and assistance to students and staff from faculty groups for the use of the vehicles; there are 5-7 research teams currently using the platforms;
• Diagnosis and resolution of functional issues and hardware/software problems,
• Installation and testing of functional modules, such as components of middleware, Autoware [1.1] and Baidu Apollo [1.2].

References

[1.1] https://github.com/CPFL/Autoware

[1.2] https://github.com/ApolloAuto/apollo

(2) Activities for Experimental Vehicle Platform Hyundai Genesis (Vijay Govindarajan, Xu Shen)

We plan to enhance and upgrade the Hyundai Genesis vehicle to provide better perception capabilities and better computing reliability, with additional sensor and hardware computing platform, as well electrical system enhancements. The Genesis platform is managed and maintained separately by Prof. Francesco Borrelli’s group.

(3) Pedestrian Data Set and Analysis (Xiao Zhou, Pin Wang)

In this project, we focus on the processing of the mid-block crossing data set collected on Hearst Street on the north side of Berkeley campus, where pedestrians often spontaneously walk across. In 2019, we plan the following activities:

• annotation of head and body pose cases to establish a meaningful training-testing data set.

method to detect head pose changing in low resolution images, and analysis of all captured video data to obtain more head-pose pedestrian-head pose cases.

BDD Vehicle Platform Instrumentation