Optimization And Implementation Of Indoor Pedestrian Positioning Algorithm Based On Monocular Vision In Occlusion Environment

Location is the information that people pay attention to and use in daily life.Global Positioning System(GPS)can not provide effective positioning services in the indoor environment.Therefore,how to provide an accurate indoor positioning method has become a research hotspot.In many indoor positioning methods,computer vision positioning technology has high positioning accuracy,and has attracted much attention of scholars.This thesis focuses on the research of indoor pedestrian positioning technologies.Monocular vision localization method,which has low computational complexity,extracts the foot coordinate of pedestrian and then converts it to the location of pedestrian.But it will be difficult for this positioning method when the pedestrian's lower body is occluded.Some researchers use binocular vision to solve this problem,but the binocular vision has high computational complexity and can not meet the real-time requirements.In this thesis,an indoor pedestrian occlusion localization method based on shoulder detection and human estimation is proposed.This method estimates the position of pedestrian's foot by using the ratio between pedestrian's head and body,and then uses the estimated foot coordinates to locate the pedestrian.This method first uses the improved seed growing based shoulder detection algorithm to get the position of the pedestrian's shoulder and the height of head.This shoulder detection algorithm has good robustness and stability.Then this method calculates the ratio between pedestrian's head and body and detects pedestrian occlusion by using the ratio model value of this pedestrian.When the occlusion occurs,the proposed method uses the ratio model database to estimate pedestrian's foot position,and then locate the pedestrian through the spatial coordinate transformation of the estimated foot coordinates.The simplicity of monocular vision makes proposed method meet the real-time requirements.In addition,the proposed method can locate multiple pedestrians simultaneously by constructing the ratio model database.Experimental results show that proposed occlusion localization method can achieve the localization accuracy of 30cm and is capable of handling multiple pedestrians' localization.Real-time test results show that proposed method's average processing time of each frame with 856 × 480 resolution is 92ms on the Core i3 2.4GHz processor.The proposed method can meet real-time requirements.Compared with the binocular vision localization method with high positioning accuracy,the proposed method's positioning accuracy decreases by only 10cm,but each frame processing time reduces by 55%.The proposed occlusion method can meet the real-time requirement with high positioning accuracy and can handle multiple pedestrians' localization.Therefore,it can be applied to the actual indoor pedestrian positioning scene.