In large electrical equipment, cables play an important role in transportation and signal transmission,and measuring the bending radius of cables is an important link in ensuring the quality of the installation process. A non-contact machine vision cable bending radius measurement method is proposed to address the issues of inconvenient operation and low efficiency in traditional methods for measuring cable bending radius. Based on deep learning RGBD bimodal semantic segmentation network, cable masks are segmented, and cable feature centerlines are extracted from mask images. Then, a cable spatial feature point set is constructed, and the cable spatial feature curve is reconstructed using parameter curve algebra to reconstruct the cable spatial feature curve, Then, the cable bending radius is calculated by using the Curvilinear motion law of the particle space. To verify the practical application effect of the method proposed in this paper, a standard arc with a radius of R = 110 mm and 125 mm was used to fix the cable.The comparison between the measurement results of the manual ruler using the chord height method and the measurement results of the method proposed in this paper shows that the average error of measuring the bending radius using this method is reduced by 7. 2% and 2. 2%, respectively, compared to the manual ruler measurement, and the measurement time is reduced by 93. 8% and 92. 2%, respectively. The proposed non-contact machine vision cable bending radius measurement method algebraically reconstructs the cable spatial characteristic curve based on mask images,calculates its spatial bending radius, and the measurement results are reliable, with higher accuracy and faster speed than manual measurement. It can be applied to various occasions of cable bending radius measurement tasks.