A New Personnel Identification and Position Estimation Algorithm Using Binocular Camera

With the increased requirement of position information for indoor environment, image based indoor localization has received much attentions, since it has low cost, no electromagnetic interference and green environmental protection. In this paper, a new personnel identification and position estimation algorithm using the images from binocular camera is proposed. In the off-line phase, the face image based training data set is used for classification learning by the convolutional neural network (CNN). The personnel identification based classification model is obtained. Then, the depth information obtained from the image are used for distance based regression learning by support vector machine (SVM). The distance based regression model is obtained. In the on-line phase, when the image and its depth information are obtained from the binocular camera, the personnel identification and the distance estimation can be obtained from the personnel identification based classification model and the distance based regression model respectively. Experiment results shown that under the largest number of training data condition, the accuracy of the personnel identification can reach to 91%. Moreover, the average error for distance estimation is less than 5cm.