Research On Recognition And Detection Technology Of Automobile Fuel Tank Lid Based On Binocular Vision

With the development of science and the improvement of manufacturing industry,the demand for unmanned refueling robot is getting higher and higher.Its emergence can reduce the work pressure of service personnel,and at the same time,car owners do not need to endure the unpleasant smell of gasoline.During this epidemic period,unmanned refueling robot can effectively reduce communication,and the car owner can complete the operation of refueling the vehicle without even getting out of the car,which is efficient and safe throughout the whole process.In the future,the refueling mode of automatic refueling robot will gradually replace the self-refueling mode and the traditional refueling mode of staff.Related technologies of unmanned refueling robot mainly include control and perception technology.This paper mainly focuses on visual perception technology,and uses binocular camera to complete target recognition,distance perception and 3D reconstruction.The research contents are as follows:（1）In this paper,the improved Faster-RCNN deep learning neural network is used to detect the outer edge of the tank lid.This paper mainly conducts a comparative study on convolutional networks in Faster-RCNN,and uses Res Net101 to better extract image features.In addition,according to the shape characteristics of the outer edge of the tank cover,the strategy of generating Ro I of RPN network is improved,and softNMS is used to screen candidate boxes.In view of the inaccurate position of the target box caused by rounding in the Ro I Pooling process,Ro I Align is used to determine the position of the outer edge of the tank cover more accurately.It is proved by experiment that better detection effect can be achieved by collecting and enhancing images for training.The experimental results show that when the threshold is 0.5,the improved network accuracy（Precision）increases by 4%,the recall rate（Recall）reaches 96%,and the average precision（Average Precision,AP）reaches 98.53%.After the improvement network has better performance.（2）The improved oil tank cover detection algorithm based on ellipse detection is used for oil tank cover detection.In this paper,the detection of oil tank cover is abstracted into ellipse detection.Different from the traditional Hough transform which can only detect perfectly shaped circles and straight lines,this paper uses an efficient and fast ellipse detection method to complete the detection of oil tank cover.Aiming at the problem of repeated ellipses in ellipse detection algorithm,this paper improves the detection algorithm and proposes a clustering algorithm to get more accurate ellipse center.（3）Use binocular camera to complete the distance measurement and 3D reconstruction,and make improvements to the specific problems existing in the distance measurement process.The parallax map is generated by SGBM algorithm and AD-Census stereo matching algorithm,and the physical distance between the target point and the camera and the corresponding world coordinate system are obtained.At the same time,the transformation of parallax map to spatial point cloud is completed,and the visualization of point cloud is completed through PCL library,which can have a better perception of the three-dimensional physical shape of the target.In order to solve the problem of large range error caused by matching error,the distance measurement method of matching error points is improved,and the bilinear interpolation method is used to complete the distance measurement and improve the range measurement accuracy.The experimental results show that the unmanned refueling robot based on binocular vision proposed in this paper can better complete the identification of the outer edge of the tank cover and the tank cover.At the same time,the range measurement and 3D reconstruction can be completed with high precision,while avoiding the use of laser radar and depth camera,reducing the cost of project equipment.