Research On The Occlusion Recognition Method Based On The Distance Estimation Of The Moving Object In The Video

As an important branch of artificial intelligence,the application of computer vision covers many fields,such as security,industry,education and so on.In recent years,the application of computer vision technology in the field of automatic driving has attracted enormous attentions from industry and academia.As a basic task of computer vision,its high quality results will be a great significance.The current algorithms for object detection are mainly divided into traditional methods and methods based on deep learning.The traditional algorithms for object detection use the features designed manually and classifiers to detect objects,while the algorithms based on deep learning for object detection mainly utilize Convolutional Neural Networks to extract the features from images,and then classify it into different classes.The advantages of algorithms based on deep learning are that they use the automatically extracted features which have good robustness instead of the tedious features designed by manual works.No matter the traditional methods or the deep learning methods for object detection,most of the researches about occlusions focus on whether the detected objects are occluded or not.Although the occlusion recognition can be achieved,the occlusion relationship can not be recognized between the occluding object and the occluded object.When we study the technology of automatic driving,we also encounter the scene of moving objects occluding each other.For the above problems,the main researches of this paper are as follows:1.In order to solve the problem about occlusion relationship recognition,we propose a method of occlusion relationship recognition,which is based on the distance measurement for the moving object in video in this paper.The theoretical basis in the method is that: For the points on the same optical axis,the image formed by the points close to the optical center will block the image formed by the points far from the optical center.It is concluded that when the object is occluded,the object close to the camera will occlude the object far away from the camera.2.Aiming at the problem of mutual occlusion between moving objects in the automatic driving,a occlusion recognition system based on deep learning object detection and binocular stereo vision measurement is designed by combining the advantages of deep learning object detection,such as fast speed,strong robustness,and small errors of binocular stereo measurement in this paper.The system uses the yolov3 model and a relationship between recognition rate and occlusion based on information entropy to detect the occlusion in video frames.When the occlusion is detected,the occlusion relationship of moving objects can be recognized and the position of the occluded object can be predicted through the binocular stereo measurement technology and the method of occlusion relationship recognition proposed above.The experimental results show that this method can effectively solve the occlusion recognition problem of moving objects and has a certain real time.