3D Parameter Estimation And Clustering Algorithm Research For Vehicle Feature Points

It is an important application to analyze the vehicle target using vehicle feature points clustering in the field of machine vision. At present, the methods of feature points clustering mostly use the positional relationship between the feature points on a 2D image plane and the trajectories constraint of them. These methods can get good clustering results under the situation that the camera is set up in a high position with a broad view. Otherwise, these methods are not applicable when the vehicle is blocked.In this paper, an iterative clustering algorithm framework is proposed after studying the method of 3D parameter estimation of vehicle feature points. It is mainly completed by using the basic clustering, the detailed clustering of the feature points and merging between categories. In the basic clustering process, the preliminary classification completed by adopting the improved K-means clustering algorithm with the analysis of the inverse projection transformation of feature points. Based on the basic clustering results, the 3D parameter of the feature points can be estimated after analyzing the relationship between the inverse projection velocity and the 3D height of them. Then the detailed clustering result is achieved through excluding and redistributing feature points by matching their 3D parameter with the 3D model. The process of merging between categories is using the motion constraints between the 3D trajectory of the feature points and the final result is achieved after merging the categories.Finally, the clustering algorithm is tested in the actual traffic scene and produces stable clustering results. It gets traffic volum by counting categories passing the detection zone in a certain period of time and the accuracy rate reaches more than 95%. The clustering results can be also used to classify vehicle model via estimating 3D parameters of feature points in each category. The accuracy rate reaches more than 90%.