Research On AGV Working System Based On The Combination Of Global Vision And Local Vision

Intelligent transportation system is more and more used in all walks of life,AGV also developing rapidly,with the development of industrial modernization.AGV system is controlled by computer vision technology,which has the unique advantages of independence,flexibility,economy and so on.However,there are many problems in the current AGV vision system,such as difficult maintenance,large amount of calculation,limited detection range,and difficult dynamic obstacle detection.Based on the above problems,we will study AGV working system,which combines global vision with local vision.The key contents and main innovations of this paper are as follows:(1)In the process of AGV detection and location in the factory,the vision of single camera is limited,and the establishment of multi camera association model is complex.In this paper,an AGV detection method is proposed,which uses the strategy of precomputing splicing parameters to achieve fast video splicing,when the camera angle is fixed.In view of the technical difficulties of image registration and image fusion in the process of video splicing,the improved surf operator based on overlapping region is used for feature extraction,and then Laplacian in surf is used for feature extraction.According to the nature of the identifier,the feature points are divided into two groups and the nearest neighbor algorithm is used for accurate and fast matching;then the best suture of graphcut is obtained based on the definition of energy function,and finally the multi-resolution fusion algorithm on one side of the seam is used for image fusion.Results showed that the average processing time of each frame was between 12 ms and 15 ms,which met the requirements of video splicing in factory application environment.(2)The receptive field of traditional YOLOV3 is often larger than that of AGV,which takes up a small proportion and has sparse features in large factory scene images.This paper presents an improved SI-YOLO3 detection network.Firstly,the feature of AGV in the factory is extracted by using the network based on residual block design,and then two roll up layers are added on the basis of the main network of YOLOV3,at the same time,on the basis of sampling the feature pyramid in 2 times of step size,it cascades with the features corresponding to the scale extracted by the previous depth network,and finally select the appropriate boundary frame to get the AGV detection model by the non maximum inhibition method.The results showed that the average detection time of each image was 35.15 MS,and the accuracy and recall rate were over 99.65% and 90% respectively.It was proved that this method had high accuracy and speed,and could effectively realize the AGV real-time monitoring and cargo status identification in the factory.(3)In order to ensure the stable operation of AGV system,a conflict resolution scheme based on environmental awareness and task priority is proposed in view of the possible conflicts in the two-way path operation of AGV.At the same time,in order to avoid the collision between AGVs,this paper designs an AGV anti-collision system based on monocular vision ranging.The system installs a camera on the top of AGV to collect the front image during the AGV driving process.The collected image is logically or operationally processed through the image after the frame difference and the background difference to obtain the contour information of the moving target.Then the distance is calculated by the shadow width ranging method so as to avoid collision.