Research On Updating Method Of Semi-dynamic Target In Semantic Map Based On Patrol Vehicle Vision

With the rapid development of China's economy,people's living standards have been constantly improved.At the same time,the situation of illegal occupation of fire safety passage caused by the surplus of unnecessary daily necessities emerges in endlessly.After emergency,unnecessary casualties and economic and property losses are common.Therefore,it is quite necessary to detect the illegal occupation of fire passage in time.However,the current safety inspection is mainly based on the field investigation of relevant staff,which is not only time-consuming and laborious,but also may be missing due to human factors.In this case,combined with the increasingly mature unmanned vehicle vision technology and the rich semantic information in the environment,it has become an effective method to detect the illegal occupation of fire safety channel.Aiming at the problem of illegal occupation of fire safety passage,this paper takes the door frame structure in corridor passage as the static target,and the object with box structure as the semi dynamic target.Based on the research status of scene understanding,target detection and location and map update,this paper focuses on the research of target detection,target location and map update with scene understanding as the core.Based on this,a semi dynamic object updating method in semantic map based on scene understanding is proposed.The main work and achievements of this paper are as follows:1.A scene understanding method based on vanishing point detection is proposed.By analyzing the relationship between the vanishing point and the horizontal,vertical and deep line groups obtained in the process of vanishing point detection,the characteristic line of corridor channel is obtained.In the region of interest,the redundant feature lines except for the static and semi dynamic targets are eliminated according to the spatial constraints(such as the object is supported by the floor and can not penetrate the wall),and the feature description of semi dynamic target and static reference target is realized.Experiments show that this method can extract the features of semi dynamic and static targets.2.In order to obtain the physical size parameters of semi dynamic objects,an algorithm based on cross ratio invariance and reference information is proposed.The geometric information of two-dimensional plane can be extracted by studying the cross ratio invariance of projective transformation.In this paper,it is extended to three-dimensional geometry.By extracting the cross ratio information of the corresponding target on two perpendicular planes,the geometric parameters of the target can be obtained by cross ratio when the reference parameters are known.The experimental results show that the geometric parameters of a specific target can still be obtained through the geometric parameters of the reference object,and the measurement accuracy is more than 92%.3.In order to realize the location and update of semi dynamic objects,this paper proposes a method of updating semi dynamic objects in semantic map based on scene understanding,combining with the information parameters of static reference objects in the manual created static semantic information database.Experiments show that the positioning accuracy of semi dynamic target is within 8 cm.Each grid in the grid map used in this paper describes the area of 100 × 100 mm in the actual scene.Therefore,the error has little impact on the accuracy of all kinds of targets described in the grid map.The algorithm basically achieves the purpose of updating and expressing the semi dynamic target in the map.