- Blog
- Innovation and Trends
- Artificial Intelligence
Eyes of the Machine: Interview with a Computer Vision Expert
Logistics Matters had the privilege of interviewing Oliver Bredtmann, a key player in the innovative world of computer vision at DB SCHENKER.
Oliver has been a driving force at DB SCHENKER since 2019, contributing to a wide range of computer vision solutions. His previous experience in the automotive sector, where he focused on AI-driven natural language understanding, provided him with a strong foundation in machine learning and its practical applications.
In this interview, Logistics Matters dives into the realm of computer vision and explores Oliver’s insights into the latest advancements and real-world applications.
Logistics Matters: Could you elaborate on your role as a Data Scientist at DB SCHENKER, particularly your specialization in AI algorithms for computer vision?
Oliver Bredtmann: We are developing AI applications, which are techniques to analyze and interpret visual data from images or videos. This involves installing cameras and computing hardware, training neural networks, and developing the application software around the neural networks. In the last couple of years, we have worked on numerous computer vision applications, like object detection and segmentation, text recognition, marker recognition, and the prediction of 3D structures from 2D images.
Logistics Matters: Can you provide details on how the training processes of neural networks work?
Oliver Bredtmann: The training process involves feeding a large, labeled dataset of images into a neural network model. During training, the images are passed through the model, which adjusts its parameters (weights) to minimize the error between its predictions and the actual labels. Over time, the model learns to recognize patterns and features in new, unseen images, improving its accuracy. To give an example: If you want to detect pallets, as is the case in our PalletVision product, you need hundreds, better thousands of images where a human has put a box around every pallet in every image. By feeding images and pallet labels into the model it learns what a pallet is and will be able to recognize a pallet in unseen images.
Logistics Matters: As a Data Scientist working on PalletVision, a DB SCHENKER AI-powered warehousing computer vision system, can you break down the AI behind the tool and how it streamlines logistics operations?
Oliver Bredtmann: The AI is trained in detecting pallets with cameras mounted at the ceiling of a warehouse. The pallet detection is the foundation of a solution that tracks the movement of pallets over time. Since a video is a series of images, the tracking algorithm determines which pallet detected in one frame corresponds to which pallet detected in the next frame. This detection and tracking solution allows for counting pallets and measuring how long pallets remain in a certain area of the warehouse. With the objective of fulfilling dock-to-stock cycle time requirements, a dedicated screen in the inbound area provides clear put-away priorities to the forklift driver. It may, for example, indicate that pallets standing in line 2 of inbound area 1 need to be put away next. At the same time, a comprehensive Power BI dashboard keeps warehouse managers informed about critical inbound KPIs. This increased visibility reduces risk for contractual penalties and streamlines operational workflows.
Logistics Matters: Let’s talk about additional projects. DB SCHENKER utilizes AI-driven systems for computer vision in Land Transport: Truck Utilization and Gate Vision. Can you explain how these AI systems work and their benefits in streamlining processes?
Oliver Bredtmann: Let me talk about Truck Utilization first. After loading trucks or swap bodies, terminal workers take pictures which are needed as input for an AI algorithm we have developed in the truck utilization project. The AI can create a full 3D representation of the swap body, which we use to measure its free volume and floor space in a 3D point cloud. This way, we obtain full visibility on utilization per lane or customer without interfering with the loading process at all.
In our Gate Vision solution, we detect trucks while they are driving through the terminal gates. An AI is trained to recognize texts off images like license plates and container IDs. Once we fully integrate this data into our transport management systems, we will have detailed visibility on the location of our equipment.
Logistics Matters: Can you walk us through the collaborative process of developing a computer vision system? Focus on the interplay between human expertise and AI capabilities.
Oliver Bredtmann: Building a computer vision system requires close collaboration between developers and business experts. Together they determine what information the AI needs to provide and how it can be translated into business benefits. With a clear solution in mind, Data Scientists start installing cameras at the target site and collecting images.
With annotated data in hand, the Data Scientist’s role is to review the latest AI models to identify the most promising models for specific tasks. Through experimentation with various models, we determine the optimal models for our objectives on specific tasks. Upon deployment on-site, we begin with a model trained on a limited set of images, improving it over time by adding more and more images that are used to iteratively refine the model to make it better and better over time. A crucial step in this process involves training a model to identify people and blur them in all stored images.
Oliver’s insights into the world of computer vision at DB SCHENKER offer valuable perspectives on the transformative power of AI in the logistics industry.
