Research On Key Technologies Of Personnel Identification And Localization Using Binocular Camera

With the development of image processing technology and the popularization of video acquisition equipment,personnel identification and localization using image processing techniques has received more and more attentions.In this paper,we study the key techniques of personnel identification and localization using binocular camera.The main contributions are as follows:(1)The theoretical knowledge and hardware of binocular camera for personnel identification and localization are studied.The image-forming principle of the camera is introduced at first.Then,some machine learning approaches used for personnel identification and localization are described in detail.At last,the hardware experimental platform of binocular camera for personnel identification and localization is built to provide a solid foundation for the performance analysis of the proposed algorithm.(2)A joint personnel identification and localization using binocular camera is proposed.Based on the obtained gray image,the problem of personnel identification can be formulated as the problem of classification learning.The person identification can be achieved through the off-line learning by convolutional neural network.For another,with the obtained gray image,the problem of localization can be formulated as the problem of regression learning.After the off-line learning by the support vector machine,the final position can be estimated.Since both the gray image and depth image obtained from binocular camera are used for joint personnel identification and localization,the cost of the proposed algorithm is low.Moreover,because of machine learning technique utilization,the complexity of the proposed algorithm is low and the performance is high.The experiment results illustrated that under the largest number of training data,the accuracy of personnel identification is 91% and the average of localization error is under 5cm.(3)A robust depth image based localization algorithm using binocular camera is proposed.In the off-line phase,the depth information which is not affected by the personnel change is chosen by the depth image cropping processing at first.Then the features are extracted with the convolutional kernel.It can remove the useless information and also reduce the noise in the depth image.Thus,it can improve the efficiency of the off-line learning.At last,support vector machine is applied to regression learning and position based regression function is obtained.In the online stage,after image clipping and feature extraction,the position based regression function can be used to calculate the position.The experiment results illustrated that under the unknown personnel condition,the proposed algorithm has better localization performance than the existing algorithm.So,it is a robust localization algorithm for practical application.