Dr. Philip Lenz
Pushing computer vision for automated driving beyond borders. Creating intelligent vehicles to make road traffic safer.
“For me, these words of the fictional character Jean-Luc Picard describe my work as a researcher best: “There is a way out of every box, a solution to every puzzle; it’s just a matter of finding it.” That is what I particularly like about being a researcher, finding solutions for challenging problems and creating benefit for society.”
I am a research scientist at the Computer Vision Lab of Robert Bosch GmbH. I am interested in object detection and tracking, sensor fusion, scene understanding, and machine learning. For me, automated driving at all levels is currently the most interesting challenge in computer vision and machine learning. My daily work is to push the borders for vision-based automated driving, making algorithms more robust and thus cars safer.
- Visiting researcher in the group led by Andreas Geiger, Max Planck Institute for Intelligent Systems
- Visiting researcher in the group led by Raquel Urtasun, Toyota Technological Institute at Chicago
- Research Assistant, PhD in Computer Vision, group led by Christoph Stiller, Karlsruhe Institute of Technology
P. Lenz (2015)Efficient Min-cost Flow Tracking with Bounded Memory and Computation
- Karlsruher Institut für Technologie
P. Lenz et al. (2015)FollowMe: Efficient online min-cost flow tracking with bounded memory and computation
- P. Lenz, A. Geiger, R. Urtasun
- International Conference on Computer Vision
A. Ponz et al. (2015)Laser scanner and camera fusion for automatic obstacle classification in ADAS application, Smart Cities, Green Technologies, and Intelligent Transport Systems
- A. Ponz, C. H. Rodríguez-Garavito, F. García, P. Lenz, C. Stiller, J. M. Armingol
- Communication in Computer and Information Science, vol. 579
A. Geiger et al. (2013)Vision meets robotics: The KITTI dataset
- A. Geiger, P. Lenz, C. Stiller, R. Urtasun
- The International Journal of Robotics Research
A. Geiger et al. (2012)Are we ready for autonomous driving? The KITTI vision benchmark suite
- A. Geiger, P. Lenz, R. Urtasun
- Conference on Computer Vision and Pattern Recognition
P. Lenz et al. (2011)Sparse scene flow segmentation for moving object detection in urban environments
- P. Lenz, J. Ziegler, A. Geiger, M. Roser
- IEEE Intelligent Vehicles Symposium
M. Roser & P. Lenz (2010)Camera-based bidirectional reflectance measurement for road surface reflectivity classification
- IEEE Intelligent Vehicles Symposium
Interview with Dr. Philip Lenz
Research Scientist Automated Driving
Please tell us what fascinates you most about research.
I started working on computer vision for automated driving more than ten years ago. I am still fascinated by how many different, challenging problems you need to solve for a machine to perceive the world like a human. Every year, more and more powerful algorithms are developed and the community pushes computer vision beyond former borders. And despite the currently achieved state of the art, there are still many challenging problems left before machines really perceive their surroundings and interact with them like a human.
What makes research done at Bosch so special?
Our goal is to develop algorithms which can ultimately be deployed in a product. For me as an algorithm developer, this means we need to team up with everybody involved in the whole processing chain, starting from camera optics and ending up with the application of an automated driving function. This allows us to improve our systems for our particular use case and introduces new challenges to make algorithms powerful enough under these particular hardware constraints.
What research topics are you currently working on at Bosch?
I am working on computer vision for automated driving. There are still many challenging and unsolved scenarios. We continuously try to push borders for hardware, sensors, and perception. I am working on the perception part. For all automated driving functions, a reliable scene understanding is mandatory. This includes object detection and tracking for all kinds of road users, which can be extremely diverse in their appearance, geometry, and behavior. Furthermore, road layout and infrastructure need to be estimated and inferred with map data. Our classical and machine learning algorithms address all these topics and must finally provide a reliable scene model for downstream applications such as vehicle control.
What are the biggest scientific challenges in your field of research?
The first automated cars were developed decades ago, and driver assistance systems are available as series products. However, the next big challenge is to develop cars that can really participate in everyday traffic scenarios without such fallbacks as safety drivers. To achieve this, algorithms must prove to be robust enough to tackle real-world traffic situations. Solving these two requirements with all the powerful, modern technologies in computer vision, such as deep learning, are currently the biggest challenges to make automated driving a real-world product.
How do the results of your research become part of solutions "Invented for life"?
People all over the world spend an incredible amount of time driving cars or even standing in traffic jams. Even worse, thousands of people still die in traffic every year. Driver assistance systems and automated driving address both issues. Accidents can be reduced by, for example, earlier reaction times or overcoming human shortcomings such as drowsiness or distraction. Resources, i.e. vehicles and road infrastructure, can be used more efficiently, reducing traffic jams and thus allowing people to spend more time on the more important things of life.