Object Recognition And Trajectory Prediction Based On Binocular Vision

With the rapid spread of artificial intelligence all over the world,the technology industry has begun to turn to the direction of intelligence and digitalization.Machine vision is one of the most cutting-edge and most valuable research in the field of artificial intelligence,and object recognition and detection is a hotspot in the field of machine vision.The main research of this thesis is to use the deep learning model to perform object recognition and trajectory prediction based on the binocular vision system.In terms of target detection,it is not only to find the position of the object in the image,but also to classify the object.This requires the cooperation of two algorithms,target detection and object classification.In this thesis,the SSD deep learning neural network combined with MobileNet is used for target detection.MobileNet as a lightweight classification algorithm can quickly and accurately extract the features of the target object.SSD as a one-step target detection algorithm can efficiently output the position of the target object.The combination of the two greatly improves the accuracy of target detection.In terms of binocular vision,the binocular camera is first calibrated to obtain the internal and external parameters of the binocular camera,and then the binocular camera is corrected so that the optical axes of the cameras are parallel to each other,and then the images of the two cameras are matched by a semi-global matching algorithm.The parallax map of the two cameras can be obtained,and the parallax information and the calibration result is further reconstructed in three dimensions to obtain the three-dimensional coordinates of the target object during the motion.In terms of trajectory prediction,the recurrent neural network can effectively predict the time series model,but due to the long-term dependence problem,the performance of the cyclic neural network has been greatly limited.In order to solve this problem and accurately predict the three-dimensional coordinates of the next position of the object,the long short term memory(LSTM)is used for position prediction.Its unique "forgetting" function can effectively avoid long-term dependence and and has better performance than the ordinary recurrent neural network.