Organizers
Scott O. Kuznicki, Modern Traffic Consultants
Ken Smith, 3M Corporation
Session Description
Are we *actually* ready for automated driving on today’s roads? Can today’s pavement markings support ADAS systems let alone Level 4 automation? Buildling on our 2018 workshop, we’ll uncover the latest information regarding machine vision and sensor fusion capabilities and failure modes related to traffic control device detection. We’ll incorporate the latest applications of simultaneous location and mapping (SLAM) approaches in support of higher levels of automation. The path from lane-keeping assist and lane departure warning to lane centering is likely to require more consistent approaches to the design, selection, deployment, and maintenance of lane marking materials. Attendees will be prepared to promote research needs related to machine vision sensors, systems, and lane marking materials and design.
Goals/Objectives/Outputs
Goal 1
Provide an overview of advances in machine vision systems (especially cameras) including the fundamental metrics for sensing lane markings in a sensor fusion environment
Goal 2
Identify key failure modes of machine vision and challenges for mixed-fleet operations related especially to lane markings and roadside delineation
Goal 3
Identify key interactions between lane markings and machine vision systems, the foundational premise for needs assessment in operating design domains to Level 4
Output 1
Identify a path towards implementation of cost-effective infrastructure improvements that maximize performance benefits for humans and CAVs
Output 2
Working group for a potential NCHRP Synthesis of Practice proposal concerning performance of pavement markings for mixed fleet environments across domains
Output 3
Attendee dialog concerning the requirements, benefits, and objectives of
a reference machine vision system with defined limitations
Agenda
The workshop is designed to generate active dialogue with both infrastructure and automotive stakeholders related to traffic control devices in general, with particular attention paid to the specific use cases of machine vision systems interacting with lane markings for the purposes of obtaining lateral positioning inputs. The workshop’s organizers have obtained engagement commitments from key stakeholders, building on five years of successful and growing workshops. Audience interaction will include conventional question and answer sessions and solicited audience feedback on specific research outcomes and needs.
The workshop will be divided into four parts according to the session plan below,
addressing the theme of this year’s session, “Sensor Fusion and Lateral Positioning Using Traffic Control Devices”.
1:30 PM - 1:55 PM Welcome, Introductions, and Road Map
The Challenging Road (Scott O. Kuznicki, Modern Traffic Consultants)
1:55 PM to 3:15 PM How do Machines See? Machine Vision Systems Capabilities and Limitations
Last year’s panel on machine vision function and failure modes related to traffic control devices was extremely popular; this year’s panel will focus on lateral guidance in various operating design domains.
- The Current State of Sensors: A Supplier Perspective (Giri Venkat, On Semiconductor)
- Processing, Sensor Fusion and SLAM: Localization & Perception via the "4th Sensor" (Ro Gupta, Carmera)
- A Vehicle Perspective on the Importance of Pavement Markings for Assisted and Automated Driving (Phil Magney, VSI Labs)
- Machine Vision for Pavement Markings in Edge Conditions (Ken Smith, 3M)
- Lidar Detection and Sensor Fusion Opportunities for Pavement Markings (Adam Pike, TTI)
3:15 PM to 3:30 PM Break
3:30 PM to 4:15 PM Developing a Reference Machine Vision System (RMVS) for ADAS Function Testing
Conducting a Safety Through Disruption (Safe-D) University Transportation Center Project Supporting Development of an Open-Source Reference Test System. (Sivakumar Rathinam, Texas A&M)
Moderated audience discussion of approach and benefits for RMVS
4:15 PM to 5:15 PM Sensor fusion opportunities to address failure modes and edge cases
This panel is devoted to understanding those situations that are most challenging for MV systems (and human drivers) including night, night-wet, snow, and glare conditions, illustrating failure modes in various ODDs.
- Giri Venkat, Ro Gupta, Phil Magney, Adam Pike, Scott Kuznicki
- Moderated by Ken Smith
5:15 PM - 5:30 PM Correlating Machine Vision Performance and Operating Design Domain
Moderated by Scott Kuznicki with input from Dr. Rathinam
Related Sessions
This session complements some of the issues anticipated in the proposed session “Public Sector Fundamental Barriers to Implantation and Deployment” organized by Ted Bailey from the Washington State Department of Transportation.
The specific goals of this session, machine vision and sensor interactions with existing infrastructure (pavement markings, in particular) are not addressed in other sessions related to automated vehicles and infrastructure needs. Emphasizing the influence of geometric design on lateral guidance and the need for improved pavement marking maintenance in some countries aids infrastructure owner/operators in planning for medium-term and long-term funding needs.
This session supports an ancillary meeting on infrastructure readiness, part of an FHWA project assessing infrastructure needs for autonomous vehicle operations. Impacts of Automated Vehicles on Highway Infrastructure Needs is being organized by Paul Carlson.