【摘要】 为克服基于视觉的转角测量方法容易受到系统干扰的局限性，提出了一种基于深度学习的机理模型和数据混合驱动的视觉转角测量方法。从数学原理上验证了采用等腰三角形作为轴上花纹的合理性和有效性，构建三角花纹转角计算机理数学模型。引入基于YOLOv8的深度学习模型，采用线性组合将两者结合构建成混合转角测量模型。实验结果显示，这种混合模型在测量准度上有显著提升，相比仅用机理模型，其平均误差降低1.125°,均方根误差降低10.05°,在不同环境测试集上仍保持高效性能。该模型充分利用了深度学习模型对图像随机干扰的学习能力，同时保持了数学模型的约束和稳定性，提高了视觉角度测量的准确性，而且增强了其对环境变化以及系统干扰的适应性。更多还原

【Abstract】 To overcome the limitations of vision-based angle measurement methods, which are susceptible to system disturbances, this paper proposed a novel vision-based angle measurement approach, integrating a deep learning mechanism and data-driven model.This study validated the use of an isosceles triangle pattern on the axis for its effectiveness and rationality, establishing a mathematical model for calculating the angle based on the triangle pattern.This paper introduced a deep learning model based on YOLOv8.A hybrid angle measurement model was constructed by using linear combination..Experimental results demonstrate significant improvements in measurement accuracy with this hybrid model.Compared to using only principle-based model, the average error is reduced by 1.125°,and the root mean square error decreases by 10.05°,maintaining high performance across various environmental test sets.This model effectively leverages the deep learning model’s ability to adapt to random image disturbances, while retaining the constraints and stability of traditional mathematical models.The precision of visual angle measurements is improved and the adaptability to environmental changes and system disturbances is boosted.更多还原