Label Detection &
Stock counting is accomplished by labels on the pallets and computer vision techniques are taken into account for this purpose. There can be multiple labels on the same pallet and it is located in different places on the surface. Therefore a camera grabs the whole area and finds a specific label using AI models. Then, text detection and recognition models are used to find text on the label, because barcodes can be really small and cannot be read in the large FOV. All three AI models are run on the drone with lots of optimizations to be able to run with localization and flight algorithms.
Python / Tensorflow / ONNX / OpenCV
The surface of the pallet can be challenging because it consists of multiple similar labels so the stock counting operation requires the identification of the correct label.
Text Detection / Recognition
Finding and interpretation of sequential data needs too much computational effort and the drone has limited resources, therefore, state-of-the-art methods must be optimized hardware specific.