Object Detection And Tracking Technology Research In Intersection Complex Scenes

Intersection is a place where two or two more roads cross, it is the strategic passage of traffic. Vehicles and pedestrians pooled, steering and evacuation there. Acquiring object moving data in intersections are of the greatest importance in transportation design and management. Because intersection usually contains many objects, and these objects have many conflicts, the traffic parameter acquiring become difficult in intersection video. Aiming at the characteristics of the intersection video, this paper discussed object detection, tracking and classifying technology in complex scenes. And a lot of algorithm and technology were presented. The main contributions of this paper were concluded as follows:(1)In moving object detection field, a spatial-temporal dual Gaussian mixture background model was proposed by extending traditional Gaussian mixture model to spatial domain. Then a double-threshold method was introduced to background detection for the spatial-temporal dual Gaussian mixture background model. By effective using of road pixel information, detection quality was improved.(2)In moving object tracking field, aiming at the characteristic of intersection, which has many objects and serious conflicts, a multi-object tracking algorithm framework was applied. By using object moving characteristic analysis to filter new tack chain generation, using MeanShift algorithm to solve vehicle conflict and using improved filter algorithm to solve global cover problem, the tracking algorithm’s stability and robustness was improved.(3)In object classification and parameter acquiring field, a trusted metrics based on object dynamic characteristic was proposed. By using the trusted metrics, a K-means algorithm was introduced for object classification. Experimental result shows the algorithm successfully classify the objects. In the meantime, imaging transform and traffic parameter acquiring in intersection video was discussed in this paper.(4)In order to acquire traffic parameter in the real traffic video based on the theory research. A general purpose data acquiring soft was developed.