AI-enabled loading documentation increases efficiency in the logistics process

Automated loading documentation during live operations

At the heart of the application is a forklift truck, with the camera mounted on its cabin. While the vehicle is in motion, the camera continuously captures the area in front of it. The process begins with recording the delivery note and covers every step of the loading sequence. This includes manoeuvring, picking up and setting down each individual pallet and finally securing the load after the forklift has been parked. For drivers, this results in a significant reduction in workload: The interruptions previously required for manual documentation are eliminated entirely. At the same time, the system generates consistent and traceable image data, creating full transparency throughout the whole process.

Intelligent image analysis in real time

A key component of the solution is the intelligent evaluation of the captured image data. Via a standardised USB Video Class (UVC) interface, the camera transmits images directly to an on-board vehicle terminal without requiring additional drivers. From there, the data is sent to a local AI server, where it is analysed in real time. First, an AI-based image classification evaluates each frame in terms of content and relevance. More than 99 per cent of the captured images are filtered out at this stage, ensuring that only genuinely relevant information is processed further. For the remaining data, specialised models from DENKweit's DENKnet system are employed. A combination of object recognition and optical character recognition (OCR) identifies delivery notes and pallet structures. The tour number is read out and serves as the central reference for assigning the images. This enables all relevant images to be linked to a specific tour and stored centrally in a structured manner. At the same time, the system can determine how many pallets appear in an image and identify their dimensions.

High image quality at changing distances

The flexibility of the solution is particularly evident in its dynamic image capture capabilities. The uEye XC autofocus camera automatically adjusts between different focal distances. “For capturing a delivery note, the focus is set to a distance of around 35 centimetres. For documenting the loading process, however, the field of view is between 2 and 3 metres,” explains Thomas Fuchs, Team Leader AIoT at CANCOM. “This ensures sharp, usable images in every situation, even when lighting conditions vary.”

The camera used features a 13-megapixel sensor and offers functions such as automatic white balance, colour correction and individually adjustable parameters for brightness, contrast and sharpness. Thanks to the UVC interface, it can be operated like a standard webcam without additional drivers. Combined with Windows IoT on the vehicle terminals, this significantly simplifies integration. Devices are detected automatically and can be managed centrally through a device management system.

Scalable AI infrastructure from edge to data centre

The particular strength of the application lies in the interaction of camera and artificial intelligence. The solution implemented by CANCOM uses containerised AI models from DENKweit. They can be scaled flexibly from the edge to the data centre and operated in Kubernetes environments, with GPU acceleration available where required. Initial working models can be created using a relatively small number of training images. However, their performance improves significantly as data quality and diversity increase.

Compared with traditional machine vision systems, this approach greatly reduces development effort. Complex rule-based vision pipelines are replaced by learning-based methods that can be implemented more quickly and adapted more easily to changing requirements. At the same time, users benefit from the rapid advances in available computing power, so even integrated GPU solutions are sufficient for many applications.