基于分层神经网络的微装配全局—局部视觉伺服研究

STUDY OF GLOBAL-LOCAL VISUAL SERVOING FOR MICROASSEMBLY BASED ON HCMAC NEURAL NETWORK

采用多视觉系统进行全局-局部3D点的跟踪,控制全局跟踪的视觉系统具有低的放大倍数和大的视场,而控制局部跟踪的视觉系统具有高的放大倍数和小的视场。在视觉伺服控制中,利用推导出的视觉雅可比映射矩阵建立任务空间到视觉空间的伺服控制方程,并利用这一方程建立预测器,对物体图像的未来位置进行预测,这样就减小了图像处理的区域,提高了图像处理的速度;同时,利用分层神经网络代替视觉空间到任务空间的映射,避免了复杂的逆矩阵计算。从仿真结果来看,这些方法的使用提高了图像处理的速度和跟踪精度。

The 3D tracking problem based on multi visual sensor systems is dealt with , of which one sensor that has low resolution and large field of view is responsible for the control of the global tracking and the other sensor, which controls the local tracking, has low resolution and small field of view .A visual Jacobian mapping from task space to vision space is developed to establish the visual servoing control equation that is used to predict the future location of the imaged object. By doing that the region of image to be processed is reduced and the processing speed is improved . Meanwhile a HCMAC neural network is fabricated to replace the complex calculation of the inverse Jacobian mapping. Simulation results show that the speed of image processing and the precision of tracking are improved by these measures.