Research Of Moving Target Recognition And Trajectory Prediction Based On Image Processing

In recent years,the researches of target recognition and trajectory prediction in the field of computer vision are developing and applying continuously.As far as video images are concerned,they can not only be used to effectively identify target objects,but also locate moving targets and predict the possible trajectories of the targets in the future by capturing the information in time series and fully mining internal features after image processing.Video images are more meaningful and valuable than the static targets in single images,which also have wider range of application.In this paper,based on video images processing,the related issues of moving target recognition and trajectory prediction are studied mainly.The specific research contents are as follows:Firstly,the convolutional neural network model in deep learning is used to recognize the processed video image data set.In order to avoid the convolution neural network training falling into the local optimal state,which makes the overall efficiency lower,and to solve the problems of slow convergence and long optimal solution time in the original genetic algorithm,an optimization recognition method based on CNN-IAGA is proposed.This method improves the population fitness of genetic algorithm and optimizes the network parameters of convolutional neural network by combining adaptive genetic algorithm.The experimental results show that the method achieves 92% recognition accuracy on the selected data set,and acquires better effect of image recognition than other recognition methods.Secondly,a new method of target detection and feature extraction(TD-FE)is proposed to detect and locate moving targets.This method consists of main body detection and detail detection.After detection,target region marking is needed to extract feature information from the final foreground images.Compared with other target detection methods,this method can suppress noise pollution and extract more clear and complete targets,it improves the foreground detection rate.In addition,an image processing system based on TD-FE is designed and implemented,which can perform image processing tasks more conveniently in this paper.Then,the target feature detection in the foreground images is quantified.The geometric and motion features of the extracted moving targets are taken as the fusion features after quantization.A trajectory prediction method based on image feature fusion(TP-IFF)is proposed.The TP-IFF method studies Elman neural network,combines Kalman filtering algorithm and particle swarm optimization algorithm,and uses particle swarm optimization algorithm to optimize Elman neural network parameters and random noises in Kalman filtering algorithm.The target fusion features are used as input parameters of Elman neural network.Through a large amount of network training,the better predictive locations of moving targets are finally obtained,and it can track targets effectively and stably.Two experiments on the selected data set show that the prediction error of this method is significantly less than that of other prediction methods.Finally,according to combine the TP-IFF method with the scenario application in intelligent transportation system,an innovative vehicle collision warning mechanism based on TP-IFF is proposed in the traffic field.