What is automotive damage AI, how does it work, and why should I care? If you are interested in these questions, stick around, this blog is for you!
How AI Damage detection works in Automotive Settings.
Wondle uses Artificial intelligence-powered technology to identify damages on vehicles from images. This blog looks at how it works, what makes it so good and how Wondle uses the data to provide the real value to the Automotive sector; actionable insights.
The basics of an AI Vehicle Damage Detection App.
Artificial Intelligence, Machine Learning and Computer Vision are all related and connected terms. They have lots in common, but central to them all is learning. This is how the platform starts to ‘recognise’ what it sees. Through teaching the model by supervising the processing of thousands, millions, and even billions of images in some models. These processes examine the images at a pixel level; this is an incredibly intensive computing process under permanent enhancement. With each process, the model becomes more accurate until the output from the model is of sufficient quality to be used in applications. So essentially, the stages of Machine learning are analysing the data, finding patterns, making predictions and learning from the feedback. That’s how the computer vision platform in Wondle was built to detect damage on leased, rental and subscription vehicles.
The model Wondle uses has been trained on tens of millions of images and is constantly refined to ensure accuracy on the two key measures, Precision and Recall. Briefly, Recall is related to identifying the presence of damage, and Precision is related to naming those detected damages correctly. If you can confidently find over 95% of damage and then confidently correctly identify what that damage is over 95% of the time, then you have a tool which is about as accurate as a trained inspector and much more accurate than an untrained member of the public.
Wondle’s Lease Return Tool with AI Damage Detection
AI Models are only as good as the images of damaged vehicles that are provided to it and so the Wondle platform ensures the quality of the images are of an engineering standard. It guides the user on each image capture and will challenge the user on any image it considers insufficient quality. It allows for 360 degrees orientation images and also for damage-specific photos. Of course, this technology is not without limitations. It cannot work in the dark or if the vehicle cannot be seen correctly because of dirt or snow, for example. It usually works quite well in the wet but water can distort light and hide very minor scratches. Splashes of dirt, especially when dried on, can sometimes be mistaken for scratches (even by a human) so it’s always best to operate on a vehicle which has been recently cleaned.
Next is the annotation of the images. The images are immediately pushed through the model returning almost instant results of the scan. This includes the location of the damage, the size of the damage, and the name of the damage (Our model uses over 40 different classifications, they are along the lines of scratch, chip, dent, crack, tear, missing part etc.). Wondle then combines this with the location to provide damage and location-specific descriptions. So, for example, you start to see Dented Rear Bumper, Scratched Left Door, Missing Right Wing Mirror etc.
How does this help the damaged lease/rental vehicle process?
Next is where the interesting turns into the essential. What do you do with all this raw data about vehicle damages? Part of the answer to this relies on another key question, ‘How bad is the vehicle damage?’ The artificial intelligence model we utilise assesses this question to give a severity score and augments this score with a confidence rating. This allows for repair assumptions and customer-side penalty calculations to be made.
As Wondle has been built with the Mobility market need firmly in focus, we have trained and developed our platform on the impact of minor and cosmetic damages. This ensures that the full complexity of decision-making involved in the processes within the Mobility space is reflected in the algorithms. This complexity is well understood by Wondle, which has been founded and managed by Leasing, Rental, Fleet Management and MAAS industry experts. Wondle utilises over 60 years of industry experience to design and develop a solution which fits perfectly with every aspect of the Mobility supply chain.
