Observing and understanding the world around comes natural to us human beings. We are able to observe and process information at an incredible rate. However, humans are prone to error and are resource extensive. Recent advances in the field of computer vision and deep learning are the driving factor for prediction models that can sometimes even outperform their human counterpart!
At Intrador, we embrace these innovations and try to think ahead;
We develop robust deep learning algorithms that help extract valuable insights for our customers out of simple pictures. Our algorithms assist decision making units in their business & logistic processes. By delegating specific tasks to state-of-the-art algorithms, we are able to drastically decrease the required human involvement in your business-process lifecycle.
Our AI models feature custom designed deep learning model architectures that are fine-tuned and optimised for their respective use-case. We closely follow innovative publications and open-source projects in the computer vision field and incorporate the here-from derived insights into our tailored solutions in an agile way.
We aspire to become the leading data broker in the asset-finance industry by 2022 and the development of object detection and classification technologies enables us to, not only solve (client’s) computer-vision use-cases, but also accelerate our own software products.
From a single picture, we can derive a great deal of information about an asset: its sector, group, make and type. But also, more complex information that usually requires a human domain-expert, such as:_ “where on the Crawler Excavator is the sprocket”_ and subsequently: “what is the condition of the sprocket?”.
This is an example of one of our custom-developed AI models that provides warnings to the authoritative decision-making units when potential fraud is detected during the asset inspections process in our mobile app.
An inspection of an asset should always be done on-location, having the asset physically present. If the asset is actually not on the location where it is supposed to be, potential fraud may be ongoing (e.g. the asset was stolen, double rented etc.)
Some inspectors were seen to not take pictures of the actual object on-location, but rather take a picture of an older picture of the asset via their computer monitors. This AI model detects whether an image was taken from a screen and effectively counteracts this type of fraud.
Intrador is experimenting with localizing and classifying damages based on pictures of the asset’s interior and exterior. Currently, we are developing use-cases for crawler excavators and tractors. Our models are able to effectively identify cosmetic damages such as rust, dents and scratches.
With every minute, our databases grow larger. We actively collect real time information of the used equipment market. Subsequently we aggregate and correlate this data which allows us to generate accurate market insights for our customers.
Wondering if you are paying/getting fair value for your remarketed machines? Our price-predictions models assist you in your decision making processes by looking at machines with similar specifications and parameters.
For a diverse set of use-cases we employ image-to-text technology (OCR) which speeds up the validation processes for our clients.
The visual inspections that are done through our mobile app usually involve assets that have a unique identifying serial/chassis-number. In light of quality ensurance and even potential fraud, its important for our customers that the claimed asset in the inspection actually matches the asset that is to be inspected. Automatic serial-number matching from pictures taken of the serial-number or chasis plate, not only speeds up the validation process, but also provides our customers with a degree of safety and quality assurance.
Besides serial-number recognition, we use OCR technoloy to automatically extract meta-data out of pictures taken from documents. With our app, an inspector saves time by just taking a picture of a legal document after which text is extracted and automaticly filled to its corresponding text field.