Research On Object Detection And Tracking Algorithm Based On Deep Feature Learning

As one of the most important research projects in the field of computer vision,object detection and tracking have a wide range of applications in various field of modern society.In recent years,some achievements have been obtained in the simple application scenarios.However,it is still a difficult thing to achieve efficient robust target detection and tracking system,because of the shape change of pedestrians object and complex environments,such as occlusion,illumination variation.Feature selection is one of the important factors which affects the target detection and tracking accuracy.At present,most of the hand-crafted features should be designed according to different data and tasks,not only need the knowledge of the professional field,but the effect is not satisfactory.The emergence of deep learning provides a good solution for this problem.Deep learning can automatically learn from the data and represent the deeper nature of the characteristics of the data.Compared with the hand-crafted features and shallow features,it has strong data description and generalization ability.In this paper,deep learning is introduced into the object detection and tracking algorithm and algorithm of object detection and tracking based on deep feature learning is proposed.The main research contents of this paper are as follows:(1)Aiming at the problem that calculation of traditional deep learning model is vast,adjusting the parameters is complex and parameter tuning needs long time,this paper proposes that a simple deep learning named PCANet is brought into extract features.PCANet is a simple deep learning model with the very basic data processing components.In this model,PCA is employed to lern multistage filter banks,then binary hashing and histograms is used for indexi ng and pooling to achieve high-level expression of the original image targets.(2)Aiming at the problem of missed detection and error detection in the traditional multi-target pedestrian detection,on the basis of extracting the depth feature of target by PCANet network,this paper adopts the selective search algorithm to extract the target window to solve the problem that sliding window used in the traditional multi-pedestrian target detection has large amount of computation and redundant windows.The experimental results show that the proposed algorithm could detect the pedestrian target accurately in the background of pedestrian target change and complex background.(3)Aiming at the problem that the existing target tracking algorithm is prone to tracking drift and even lose the tracking target in the complex environment,in this paper,a method of combining particle filter and detector is used to implement an online target tracking algorithm.In this algorithm,particle filter as the motion estimation obtains the new detection region samples.Then the feature of the samples is extracted.The SVM classifier is used to classify the feature.When the target is changed greatly,the network and the classifier can be updated by setting the threshold value.The simulation results show that the algorithm proposed in this paper has good tracking ability in the video sequences with complex background such as illumination change and occlusion.