A Design And Implementation Of 3D Robot Vision Sensor

With the development of robot technology and its wide application, the robot needs better vision. Especially for mobile robots, the ability of using 3D vision sensors to create a map of directly determines the degree of robot’s intelligence. However, the expensive price limits the wide use of 3D vision sensors in the robot for scientific researching, Business applications and civilian applications.Based on the principle of distance measurement in laser triangulation, a 3D-line-laser vision sensor was developed with low cost and high scalability, using microcontroller named Arduino due, HC-05 Bluetooth serial port, SG90 steering gear and PTC video capture chip, of which the measurement accuracy and scanning speed can be controlled. The entire sensor system consists of a laser optical measurement equipment, thread scheduling and control logic unit, Bluetooth and serial communication module, an image processing algorithm unit and point cloud, the mechanical structure and electrical transmission equipment and MeshLab PC and other components.Firstly, the paper designs the line-laser-scanning model based on the principle of distance measurement in laser triangulation, so that a single photo shotted by the camera can measure the distance between each sample point on a straight line and the sensor simultaneously. Then it goes to design the data collection unit and logic processing unit of the host controller, which obtains real-time video information from the high-speed camera UART and receives the measure format send from PC via using Bluetooth to control the speed and angle of the motor. Next, a jpeglib library is realized on the Arduino, and the JPEG compression pixel data is converted into BMP format data. After smoothing and filtering process, the sensor obtains a set of point cloud array by extracting the target feature through Gaussian fitting method and Fourier phase legal translation.The cloud data on motor rotation angle of each time point is compressed into Octrees octree format through triangulation algorithm, and is generated to K-D tree. Finally, the cloud point data is send to Labwindows and displayed on the host computer MeshLab through Bluetooth serial port, ranging results of the three-dimensional reconstruction. The ranging results are reconstructed.According to the result of the distance calibration experiment and angle calibration experiment, the parameters and component type of the system mechanical structure are constantly adjusted and optimized, in order to demonstrate and analysis the measurement error. Experimental results show that the designed 3D laser sensors can collect data quickly and flexibility by comparing with the normal laser sensor from the cost, size and other aspects, which is an effective program for obtaining high-quality three-dimensional data.