Research On Three Dimensional Positioning Technology Based On Embedded Vision

The optical positioning system based on multi-vision vision has the advantages of wide measuring range and high positioning accuracy.Widely used in industrial,medical,aerospace and other fields,such as assembly robots used in the industry,surgical robots in the medical industry have high requirements for the positioning accuracy of moving parts.In order to make the optical positioning system have higher positioning accuracy,this paper uses the machine vision technology to build a four-eye camera positioning system to achieve high-precision dynamic detection of the three-dimensional position of the moving target.The near-infrared illumination circuit,infrared light marking points and filters are used in the image acquisition part,which effectively eliminates the influence of visible light in the environment,reduces the complexity of image processing,and improves the robustness of the system.Select the pinhole model as the basic model of the camera.The improved Zhang Zhengyou calibration method is used in the camera calibration process to make it suitable for camera calibration under infrared light conditions.In the 3D reconstruction,a 3D point reconstruction algorithm based on the weighted average of the foot-points is used to obtain the high-precision position information of the points.In order to solve the problem that the image processing process is realized on the PC,the system is bulky and not portable.This paper designs an embedded image processing system based on ZYNQ-7020 processor.Image feature extraction and high-speed data transmission.The system overcomes the shortcomings of the traditional embedded image processing system,such as large power consumption,long delay and limited bandwidth.In order to maximize the advantages of combining the processor system(PS)and programmable logic(PL)in the ZYNQ chip,a computationally intensive feature extraction algorithm is implemented in the PL,and the parallel processing of the PL part is utilized to improve the processing speed.At the same time,the PS part is used to control the processing flow of the entire system,which improves the flexibility of the system.When extracting the coordinates of the center point of the image,a custom IP core is generated using the high-level synthesis tool in Vivado.In the data transmission interface,the AXI4-stream bus is selected to transmit image data.The final image processing speed test results show that the image processing and transmission scheme used has greatly improved the speed of the system.The positioning accuracy of the four-eye camera positioning system largely depends on the accuracy of the model during the three-dimensional reconstruction process.The traditional Euclidean reconstruction method is confusing in the calibration process,the process is cumbersome,and it takes a long time for the real-time system.In response to this problem,this paper uses a reconstruction model based on radial basis function neural network(RBFNN).Excluding the previous camera calibration,conjugate point matching and other intermediate links,all parameters are obtained through the neural network training process.In addition,in order to improve the stability and accuracy of the reconstruction model based on radial basis function neural network,a special image preprocessing method is proposed.The experimental results show that the reconstruction model has good robustness in both static and dynamic conditions,and the system's three-dimensional positioning accuracy reaches 0.26 mm.