How AI locates damages in the automotive industry

In this article, I will be delving into the benefits of using AI technology for detecting damages and the way it works. 

If you are interested in finding out about MakoLa’s damage locator offering, please see our AI damage locator case study.

AI and deep learning (DL) algorithms are designed to support a broad range of fields, including:

• robotics;
• medicine;
• computer vision;
• the insurance sector;
• data analysis;
• fraud detection;
• risk mitigation;
• automating claims processes.

Leveraging machine learning (ML), computer vision and neural networks enables AI systems to analyse images, video material and sensor data, providing precision detection of anomalies or damage. This makes AI useful to the automotive industry in a number of areas, as set out below.

Manufacturers are increasingly relying on AI to ensure that their vehicles meet the rigorous quality standards required of them. AI-powered systems are capable of detecting faults and pinpointing scratches, dents and irregularities during the production processes. They can also monitor the processes and analyse data in real time in order to locate flaws.

Predictive analytics grounded in historical data is used to predict potential failure points. This type of system warns technicians about problems rapidly. Catching them before they get worse means that costs are reduced.

AI-driven systems facilitate faster assessments during claims handling and repair processes. They provide quicker damage recognition, analysing vehicles after accidents in order to estimate repair costs. Insurers, on the other hand, use AI to streamline their claims approval processes.

The damage locator is a breakthrough solution in the vehicle insurance sector, minimising the problems connected with damage assessment and reducing the costs of handling claims. In the face of the strict regulations in place, AI technology, including deep learning, greatly enhances manual vehicle inspection processes. With advanced detectors such as You Only Look Once (YOLO) v5 and YOLOv8, damage locators have the capability of precisely identifying and classifying damage to vehicles on the basis of photos. 

These damage locators use data sets prepared by experts and insurers, which makes it possible to obtain high-precision damage assessments. That, in turn, speeds up the claims handling process significantly. Insurers can leverage AI technologies not only to increase their effectiveness, but also to reduce losses connected with claims leakage, which is the difference between what a company actually spends on settling a claim and what they should spend. This is estimated as being around eighteen billion dollars a year in the United States alone. 

The vehicle insurance sector has seen deep learning integration being used in damage assessment more frequently of late. This comes as no surprise, given that the development of automated damage assessment technology has proved to be highly beneficial, demanding less manual work and providing much more damage inspection efficiency. During routine claim processing, insurers can save a great deal of time and money thanks to the automatic identification of minor exterior damage like broken glass.

Detecting vehicles by way of traditional image processing methods is based on visual analysis and the use of algorithms enabling the identification and tracking of objects. These methods often employ techniques like the Canny edge detector, shape recognition and the Hough transform for identifying vehicle contours. In addition, image segmentation techniques are used, making it possible to separate vehicles from the background by means such as colour thresholds or texture analysis, for example. Although traditional approaches produce satisfactory results under controlled conditions, their efficacy can deteriorate in less favourable circumstances, such as poorly lit spots, for instance. This is why more advanced solutions rooted in machine learning and artificial intelligence are now being used with growing frequency.

Detecting vehicle damage through the use of deep learning algorithms is a cutting-edge method which has revolutionised image analysis and object detection. One of the most popular methods uses neural networks such as convolutional neural networks (CNNs), which are capable of automatically extracting important features from images, significantly improving detection accuracy. YOLO and Single Shot MultiBox Detector (SSD) models, on the hand, facilitate simultaneous damage detection and location. 

Deep learning algorithms also make it possible to train models on large data sets, meaning that systems can be better adapted to a wide range of lighting conditions, viewing angles and vehicle types.

Artificial intelligence, and CNN-based techniques in particular, play a crucial role in performing image analysis in damage locator technology for the automotive industry. Algorithms like YOLO and SSD are widely used for identifying and locating defects in vehicles.

CNNs are a fundamental element of AI-driven image analysis. These neural networks are designed with a view to processing spatial data such as images. In the context of damage location, they are used to:

• detect defects, in other words, to identify cracks, dents and chips in, or scratches on, the surface of bodywork;
• categorise damages, which involves classifying defect types, such as a ‘scratch’ versus A ‘dent’;
• locate damages by determining their coordinates in the photo under analysis. 

CNNs work through three stages during damage location.

1. The convolutional layer, where the features of the image, such as edges, textures and surface patterns, are segmented, helping to identify specific characteristics like irregularities in paintwork.
2. The pooling layer, which reduces the dimensionality of the data, retaining the most important information. This, in turn, increases the computational efficiency.
3. The fully connected layer, where the extracted characteristics are combined for classification purposes, such as ‘Dent, X mm deep’, for example.

YOLO is an object-detection algorithm that analyses images in real time, dividing them into a grid and predicting object classes and their location (bounding boxes) at the same time. 

The YOLO algorithm is popular in the automotive industry on account of:

• its speed; it can process images in real time, which is vital to work on a production line, for instance;
• its uniform approach; it simultaneously identifies damages and determines the sites.

YOLO performs several damage location activities.

1. Division of the image: the image is divided into a grid (13x13, for example) and every cell in the grid predicts a bounding box and the probability that a defect is present.
2. Predicting the object classes: YOLO predicts the class of damage, such as ‘crack’ or ‘scratch’.
3. Non-maximum suppression (NMS): a technique for eliminating non-essential or overlapping bounding boxes around the same object. The process improves detection results, ensuring that every object is represented by one bounding box with the highest probability. To give an example, a system analyses a photo of a vehicle’s bodywork and YOLO marks off the damaged areas with bounding boxes and descriptions, such as ‘Dent, 85% certainty’. 

SSD is another fast and precise algorithm object-detection algorithm. It operates similarly to YOLO, but has a more complex bounding box prediction system and is used in the automotive industry because it provides analyses at various degrees of detail. It generates numerous proposals (bounding boxes) for potential damages, with every proposal evaluated in terms of the probability of damage and its classification, and it has the advantage of being able to identify large dents and small cracks simultaneously in a single image.

SSD also offers:

• multiple feature maps; the algorithm uses various levels of image detail, facilitating the identification of major and minor defects;
• fast operations; like YOLO, SSD is optimised to operate in real time, which is crucial in automated quality control systems.

Integrating AI into damage detection processes provides numerous advantages. 

1. Speed and efficiency: AI reduces the time required to check vehicles. Traditional manual inspections can take several hours, or even days, while AI systems can perform the task in just a few minutes.
2. Precision: AI is capable of analysing data consistently, which eliminates human error and ensures extremely accurate results. This is particularly vital when it comes to identifying hidden or minor damages and detecting defects in spots that are hard to access.
3. Cost savings: the large databases of damage locator systems mean that they can carry out damage recognition rapidly and efficiently. This reduces repair costs and operating time. 
4. Scalability: AI systems can handle large quantities of data.
5. Better client/customer experience: AI improves the client and customer experience by speeding up the insurance claims process.

Despite its advantages, AI faces several challenges in the automotive industry.

1. Data constraints: the AI systems are scalable, which is a great advantage. However, effective damage locators need enormous quantities of data that have been correctly segregated, which can be time-consuming and costly. The systems require large, high-quality data sets for training.
2. Adaptation problems: something that should be remembered with technology that leverages AI is that it can still create problems. Vehicles differ in terms of size, shape and materials, which might cause errors on the part of a damage locator. AI models have to be adapted to different vehicle types and that, again, demands time-consuming training.
3. Ethical concerns: in an era of multiple applicable laws, there will always be concerns about data privacy and potential errors in AI algorithms, particularly in the context of insurance.

The integration of artificial intelligence and vehicle damage detection will continue to develop. AI systems will increasingly combine visual data and other inputs like sensor readings in order to provide a holistic picture of a vehicle’s condition. AI-powered robots could eventually carry out minor repairs, reducing the need for human intervention. AI also makes it possible to detect vehicle damage in real time, which enhances safety. Moreover, in all likelihood, insurance companies will accept AI-detected damages on account of the simplified processes and savings they give rise to.

Artificial intelligence is transforming the automotive industry, enabling fast, accurate vehicle damage detection. Deep learning algorithms like CNNs, YOLO and SSDs are opening the door to automating quality control, maintenance and damage assessment processes. AI technology is increasing efficiency, minimising human errors and reducing costs in both the automotive and insurance sectors. It is therefore not in the least surprising to see the trend for investing in AI integration in vehicle diagnostics and repair continuing to surge. 

1. BMW Group, AI in Quality Control;
2. Corzo-García D., L. Pro-Martín J., Álvarez-García J. A., Martínez-del-Amor Miguel A., Fernández-Cabrera D., Pérez-Zarate S. A., Automated Car Damage Assessment Using Computer Vision: Insurance Company Use Case, “Applied Sciences” Volume 14, Issue 20, 2024;
3. European Space Agency, Applications of AI in Industry;
4. GeeksforGeeks, Difference between YOLO and SSD;
5. Harvin H., YOLO and SSD in Deep Learning | Machine Learning Tutorial Beginners;
6. Hoang, V. D., Huynh, N. T., Tran, N., Le, K., Le, T. M. C., Selamat, A., & Nguyen, H. D., Powering AI-driven car damage identification based on VeHIDE dataset, “Journal of Information and Telecommunication”, 1–19;
7. Kumar S., From SSD to YOLO: Introduction to Modern Object Detection, Tredence
8. McKinsey & Company, AI and the Future of Automotive;
9. Qaddour J., Siddiqa S. A., Automatic damaged vehicle estimator using enhanced deep learning algorithm, “Intelligent Systems with Applications” Volume 18, 2023;
10. Tesla, Predictive Maintenance Technology.