What the damage knows that the claim form doesn’t

Why the point of impact decides who pays

In many countries the person who caused the crash must pay for the victim’s damage: medical costs, lost earnings, asset damage and suffering. They cannot claim on their own insurance, because they are at fault. Determining the point of impact and the cause of incidence is therefore what assigns liability — and liability determines who pays. As the number of vehicles on the road rises, so does the number of accidents, and so does the volume of these determinations.

Definitions used in this analysis

• Point of impact (POI): the exact or most likely point at which two or more vehicles or objects make contact and create the force of impression. The force then dissipates around that mean point.
• Scratch: damage where a panel or part experiences a force of friction, peeling the surface paint.
• Dent: damage where a portion of a panel, usually metallic, deforms on impact and creates a crater.
• Broken part: a part that falls off on impact, breaks up completely, or pops off the skeleton structure of the vehicle.
• Total loss: damage in which the entire front of the vehicle is impacted, with extensive damage to the engine and engine compartment.

The traditional approach, and why it fails

Existing practice depends on human intervention and on interpreting the written words of the claim applicant as submitted to the insurer. The source is the First Notice of Loss, which is not a reliable or objective method for determining the point of impact and cause of incidence.

The weakness is structural, not occasional.

A written account is authored by an interested party. That creates room for the aggravation of damages — presenting pre-existing damage as part of the current incident. No amount of care in reading the account removes the incentive that produced it.

The computer vision approach

POI assessment belongs to the field of advanced image processing and analysis systems — specifically computer vision using deep learning and neural networks. The system processes multiple images captured at the site of the accident, and AI algorithms determine the extent of damage across the vehicle’s exterior. The correlation between the damages detected across those different images forms the basis for predicting the cause of incidence.

The process, in five steps

• 1. Image capture. Up to eight images of the vehicle from different angles are submitted as input. Guided visual inspection through AI camera models operating in automatic mode ensures the quality of both the collected images and the inspection results.
• 2. Identify the object of interest. Computer vision extracts only the required object of interest from each captured image.
• 3. Vehicle damage analysis. Deep learning identifies and classifies damaged parts, attaching a severity rating and a damage type to each.
• 4. Grid-based analysis. A scoring system correlates damage on a single part across multiple angles, identifying the specific area within that part where the damage sits.
• 5. Point of impact aggregation. The underlying neural network aggregates the features analysed in the previous components to predict the prospective primary point of impact, and from it the potential cause of incidence.

What this changes, and what it does not

The causes of incidents are debatable by nature. They vary with circumstances that often cannot be recreated and do not exhibit recurring behaviour. No method returns certainty, and this one does not claim to. What the computer vision approach removes is a different problem: human error, and the preconceived ideas that colour judgement. The analysis runs on the vehicle’s own evidence, scored identically every time, independent of who wrote the claim.