AI-Based Bicycle Road Crack Detection

Bicycle Roads: The Risks Caused by Neglected Maintenance

Asphalt roads are damaged by repeated loads and climate changes, resulting in potholes, cracks, and other forms of deterioration. If proper maintenance is not carried out, such damage can threaten the safety of bicycle users. However, as maintenance budgets are reduced due to national financial constraints, road conditions continue to deteriorate. In fact, accidents caused by cracks and height differences between road curbs have continued to occur on bicycle roads in Hangang Park. The number of accidents reached 107 in 2022, 117 in 2023, and 88 as of September 2024. Insufficient maintenance is becoming a real safety risk.

Problems with Current Bicycle Road Maintenance

The problem with bicycle road maintenance is not limited to budget shortages. Current maintenance is carried out through manual inspections, where people directly check the road conditions and record data. This approach requires significant time and cost, while also making objective judgment difficult. Bicycle roads have basic standards such as width, color, and material, but there is a lack of clear criteria for determining the level of damage at which repairs are required. Existing maintenance standards are highly qualitative, making them ambiguous as practical management guidelines. For example, one such standard can be interpreted as follows.

Pavement condition that provides a high level of ride comfort but shows signs of surface deterioration

This is not a quantitatively clear standard. The problems mentioned above can be summarized as follows.

1. Time and cost consumption caused by manual inspections
2. Ambiguous maintenance criteria due to reliance on inspectors’ subjective judgment
3. Lack of damage classification standards based on quantitative criteria

Then, can AI technology help solve these problems?

Bicycle Road and Crack Detection Using AI

Two main capabilities are required. The first is recognizing the bicycle road itself, and the second is detecting cracks that occur on the road. The results obtained through the AI model are shown below.

As shown in the image, the AI model detects the bicycle road area and identifies cracks that have occurred on that road.

How AI Detected Bicycle Roads and Cracks

To detect bicycle roads and cracks, we built a YOLO-based deep learning model using pre-collected training data. This model was designed to identify bicycle roads and analyze whether cracks exist. The model development process is as follows.

Labeled images were used in advance, and the data was preprocessed into a format suitable for training before being used in the model. The process consists of the following steps.

1. Extract only the labeling data while preserving the original images
2. Convert the existing label data into a format suitable for model training
3. Keep only the target objects to be inferred from the label data and remove the rest
4. Train the model and verify the results

Through this process, we developed an AI model capable of detecting bicycle roads and cracks. The inference process, which refers to the process of viewing results using the trained model, is as follows.

The structure involves inputting the image to be analyzed into the trained model and then post-processing the resulting output. The internal model structure and post-processing process can be summarized as follows.

1. Extract features from image data to form a feature map

2. Calculate numerical data using processes such as NMS and IoU calculation

3. Perform post-processing based on the original image and numerical data generated by the model

   • Generate an inference result image using coordinate data and the original image
   • Or perform numerical analysis using values such as the size of the detected areas

Effects of AI-Based Bicycle Road and Crack Detection

The trained model can be used to detect bicycle roads and the cracks that occur on them. The key benefits can be summarized into the following three points.

1. Reduction in cost and time
2. Collection of objective maintenance criteria
3. Creation of safer bicycle road environments

An AI-based automated approach can reduce both cost and time. In addition, it enables quantitative evaluation through a system-based process. As a result, maintenance efficiency can be improved, making it easier to create safer bicycle road environments.

Innovating Bicycle Road Maintenance with AI Technology: Improving Efficiency and Creating Safer Mobility Environments

AI-based bicycle road and crack detection is considered a technology that can significantly improve the efficiency of bicycle road maintenance. By moving away from conventional subjective and inefficient inspection methods, it can help establish more objective and quantitative maintenance standards. This can reduce maintenance costs while lowering accident risks and providing a safer mobility environment for bicycle users. The safety of bicycle roads is not merely a matter of management, but an important issue directly connected to citizens’ quality of life. If AI technology is applied to maintenance, it will be possible to build a more sustainable and safer bicycle road environment.