Multi-feature Based Image Matching And Its Application In ITS

In computer vision,image matching is an important branch,which is widely applied in many fields such as target recognition,target tracking,satellite remote sensing,image interpretation etc.In this paper,we study the existing classic image matching algorithm at home and aboard and apply them in vehicle tracking in the field of the intelligent transportation system(ITS).Taking the complex background factors into account,such as shadow,light,pseudo interference,we give consideration to both real-time and accuracy and put forward a adaptive system of vehicle tracking in the real road environment.The main work is:(1)In the image preprocessing stage,according to the principle of Gestalt psychology,the surrounded continuous objects are tend to be targets,we use a simple and efficient saliency detection method called BMS model,take advantage of the device independence of Lab space,transform original pictures into Lab space and obtain the boolean map by continuously sampling the threshold in three channels.Then,an attention map is generated by using FloodFill algorithm in each boolean map.Linearly combine the attention maps to get the average attention map.In this way,target and background are separated to reduce the computation of the post-processing.(2)Subtracting two adjacent frames to extract horizontal edge feature from the differential image.A vehicle descriptor is established in the edge feature,avoiding large amount of global search of the feature points.It detected vehicle and eliminated many interference factors such as pedestrians,motor vehicles,shadow,and lights etc at the same time.(3)The historical correlation templates are established,SIFT feature is extracted in these who cannot match by edge feature.In the light of symmetry feature of vehicles leading to false matching,we propose a matching purification method based on the constraints of the uniformity of the adjacent points' angles to purify the matched points,which can avoid the reappeared vehicle misjudged as a new target,thus enhance tracking continuity.