Noncontact measurement of tire deformation based on computer vision and Tire-Net semantic segmentation

Tire deformation is a contributor to vehicle driving safety and an important parameter reflecting the vehicle-road/bridge interaction. Existing methods for measuring tire deformation require measuring devices in contact with the tires, which has many limitations in practice. Thus, the present study proposes a noncontact measurement method of tire deformation based on computer vision and deep learning techniques. Firstly, the diverse dataset of tire images is established based on tire images collected from roadside cameras. Next, a new semantic segmentation Tire-Net is developed to segment the tire images. Then, the quantification algorithm including the subpixel-level edge detection, key point positioning, and scale factor determination is proposed to calculate the physical value of tire deformation. Finally, field tests are carried out on the tires of cars, buses, light trucks, and heavy trucks to verify the proposed method. The results show that it performs well as a means of measuring tire deformation.