Research On Conductor Detection And Gesture Instructions Recognition Based On Monocular Vision In Moving Situation

Under the background of information warfare, computer technology represented by pattern recognition and artificial intelligence is applied in every field of military affairs. In this paper, we studied the gesture recognition technology of the standing posture command personnel in the case of mobile shooting. In order to overcome the low accuracy of hand recognition and the difficulty of identifying hand trajectory, this paper adopts the following scheme: first of all, detect the command staff to determine the location and scale of size; secondly, determine the detection area of the face and hands according to the geometric relations. Finally use the relative displacement direction feature for gesture recognition.First of all, the standing position command personnel detection algorithm was studied, through the experimental comparison of several features and the classification of the command personnel detection performance. We took the commander's leg as the detection target for command personnel detection. Aiming at the problem of high false alarm rate of the Haar feature cascade classifier, we proposed a method based on the combination of Haar feature and contour feature. The method uses Haar features and Adaboost classifier training to get a cascade classifier, and then use Hausdorff distance to generate weak classifier, and finally uses the Adaboost algorithm to get a strong classifier as the last cascade classifier.After obtaining the position of the command staff, we reduced the detection area according to the geometric relationship between the face and hand, in order to improve the detection efficiency and detection accuracy. In this paper, we use the Haar features and LBP features to do the face detection testing, and finally we decided to use LBP features for face detection and localization for the higher efficiency.In this paper, the HSV color space was used for hand segmentation. Due to interference in hand gesture segmentation based on color space, we firstly exclude skin color region by geometric relationship, then execute the ??- filtering to further suppress interference.In the part of gesture trajectory recognition, we used the relative displacement direction features to train the hidden Markov model, and used the model to do the single hand gesture recognition. for Both the relative displacement direction feature and hidden Markov model were not effective in static hand recognition, firstly we distinguished the static hand, then use the k nearest neighbor method to classify the static hand. finally, recognized the left and right hand trajectory use the hidden Markov.Good results have been achieved in the hand gesture recognition experiment.