Research On Recognition And Location Method Of Electric Vehicle Charging Port In Complex Environment

In recent years,with the rapid advancement of automobile electrification,network integration,intelligence,and sharing,the automatic driving and automatic parking functions of electric vehicles have gradually become popular,and the demand for automatic charging has become increasingly urgent.At present,the combination of robot and machine vision is the main technical route to realize automatic charging,and the recognition and location of electric vehicle charging port is one of the key technologies.In this paper,the monocular vision is combined with the six-degree-of-freedom robot system to solove the recognition and location of charging port in complex environment.For the difficulties in complex application environment,the charging port recognition method based on convolutional neural network and the charging port location method based on circular feature are studied.The main contents of the thesis are:Firstly,the accuracy requirements of the docking experiment for positioning are analyzed,according to the internal characteristics of the charging port and the difficulty in complex application environment,the charging port recognition and location scheme is designed.The visual characteristics of the charging port recognition and location are designed.An experimental verification platform based on the combination of six-degree-of-freedom robot and monocular vision was built.Zhang Zhengyou calibration method and Tsai method were used to complete the calibration of the camera distortion model and the hand-eye calibration under the eye in hand configuration mode.Then,it is difficult to identify the charging port image in the visual search process due to local missing,pseudo charging port,cover not open,illumination change and other complex application environments,proposed the charging port recognition method based on convolutional neural network.The charging port image sample set in complex application environment is established.Based on the LeNet₅ model,the learning rate and network layer number are optimized.The convolutional neural network recognition model of the charging port is constructed,and the training is completed with the sample set.The test set was classified and identified to verify the validity of the model.Then,considering the location accuracy requirements and the characteristics of the charging port,the charging port positioning method based on the circle feature is proposed,which mainly includes three steps:image preprocessing,feature detection and pose solving.Image pre-processing,including image brightness adjustment and median filtering;in the charging port feature detection,HSI color model conversion,Otsu threshold segmentation of hue components,edge detection of mathematical morphology combined with canny,improved Hough transform ellipse By fitting and other steps,the imaging ellipse and its parameter information are obtained.The geometric method is used to solve the circular pose with the known radius,and the characteristics of the charging solution are used to eliminate the ambiguity of the pose solution,and the charging position is obtained.Finally,combined with the above theory and method,the integration of the plug-in experimental system is completed,and the charging port position is converted to the robot tool coordinate system.The accuracy of the charging port recognition and positioning algorithm is verified by the plug-in experiment,and the positioning accuracy satisfies the docking.The robustness of the proposed algorithm to different illumination environments is tested and the experimental error is analyzed.