| With the increasing shortage of labor force in China and the proposal of China intelligent 2025,the development of intelligent robots has become an urgent need in China.At present,most of the robot's work is to perform repetitive specified actions through preset programs.Once the operating environment changes,the robot needs to be reprogrammed to adapt to the new environment.The robot with object detection function can be endowed with the ability to sense the environment,so as to carry out servo control according to the change of the environment.In this paper,by studying the object detector and the robot visual servo system,the hardware and software platform of the object detection robot is built,and a series of improvements are proposed for the object detector.The main research contents are as follows:Aiming at the application requirement of mobile robot in changing environment,the overall scheme including mechanical structure and hardware scheme is designed.First on the mechanical structure,the robot body structure using lightweight design of aluminum alloy plate,and the design has two layers of bearing plate,for hosting various sensors and other objects,drive mode adopts differential drive,and differential wheel attached directly to the servo motor,can be very precise and convenient to control the robot to go forward,backward and steering,robot design speed for no more than 40 m/min,detection range is designed for 0.3 to 10 m.For the hardware system of the robot,this paper USES a two-layer board card design scheme with the cooperation of upper and lower computers.PC using TX2 as main control computer,data analysis of visual sensors,and then through the CAN bus transfer results to the next bit machine,a machine use STM32 through RS232 and RS485 communication mode control response to a variety of hardware components,and selects the depth camera for visual sensors to collect images,to build the camera model,and the whole software system was designed,the whole software system CAN be divided into object detection module,robot navigation and positioning two big modules.To solve the problem that most object detectors are difficult to detect in real-time in robot embedded system,based on single-stage object detection algorithm SSD,a method to replace the trunk network and reduce the number of network model parameters is proposed to further improve the forward inference speed.Fusion at the same time put forward channel attention mechanism of SE-MSSD object detection algorithm,the algorithm using deep separable network MobileNet VGG16 network for the replacement of SSD detector,and MobileNet network structure,on the basis of the redesigned and improved,and the trunk network adopted the 17th floor with channel attention mechanism of SE network layer and layer 8 standard convolution.The improved SSD object detector was trained and verified by experiments on Pascal VOC data set,which improved the object detector's accuracy and detection speed by 69.6%.In order to verify the effectiveness of the improved SSD object detector and realize the obstacle avoidance navigation function of the robot,Realsense D435 camera was calibrated based on perspective projection model and camera distortion model.The distortion coefficient of camera's internal participation was calibrated and the calibrated result display error was within 1 pixel.The improved SSD object detector was used to identify and locate the object.In the end,the recognition confidence was more than 81%,and the detection speed was about 30FPS.Finally,the pre-trained object detection model is deployed to the robot's visual servo system,and a simple environment is built to realize the object identification,positioning and obstacle avoidance navigation of the robot in the actual environment. |
PDF Full Text Request