Automated Odometer Mileage
Extraction
United Kingdom
For an insurance company, one of the key factors is to give the customer a good experience by not asking them to fill multiple forms regarding the car details which may lead to human error and eventually may delay the process. An accurate mileage reading is a key piece of information that is relevant for auto insurance quotes and claims processing. A perfect way to get the odometer reading would be by having the image of the odometer dashboard and then processing that image through some automated image processing technique, to get final odometer mileage reading.
Client
Automotive InsurTech Company
CLIENT
The United Kingdom is one of the largest automotive innovation hubs and home to few of the emerging AI adaptive insurTech startups. Our client is one of them, with a great vision to unlock the future of insurance with innovative technologies, solving auto insurance fraud detection and claims automation, while empowering insurers to deliver amazing customer experiences.
CHALLENGE
For an insurance company, one of the key factors is to give the customer a good experience by not asking them to fill multiple forms regarding the car details which may lead to human error and eventually may delay the process.
An accurate mileage reading is a key piece of information that is relevant for auto insurance quotes and claims processing.
A perfect way to get the odometer reading would be by having the image of the odometer dashboard and then processing that image in our model to get the individual seven segment digit region and identifying each digit.
SOLUTION
We have built an object detection model to first identify the exact odometer region which was trained on the odometer regions, then after the identification of the region we have passed that particular region to the next model that we created that is the seven-segmented object detection model, where this model is trained on the seven-segmented displays of every digit from 0 to 9 as well as few characters like ‘KM/Miles’ which indicated Kilometre/Miles and ‘TRIP’ which indicate a specific trip reading but not an exact odometer reading, this model identifies all the individual digit and character from the odometer region. All the individual digits and characters are the output here.
Data Description:
Image kind data is used for our use-case. The data is collected from Google images, YouTube videos and manually clicked pictures. In the first model that is the odometer region detection model, odometer dashboard images are used to train the model where the exact odometer region is annotated and fed to the model. The second model, that is the seven segmented model, the output image from the first model that is the cropped image of the exact odometer region is used as the input for the second model.
For both the object detection models we have used pre-trained Faster RCNN Inception net architecture to train our custom model. The odometer region model is trained for 4500 epochs. The seven-segment model is trained for 15000 epochs.
RESULT
We have tested both the models on the new unseen data, the accuracy of the odometer region detection model is close to 96%, and the accuracy of the seven-segment detection model is 99.28%.