Vehicle Tracking Algorithm Under Occlusion In Video

As an important part of the ITS, object occlusion often appears in tracing and recognizesthe moving target based on image, this is a difficulty to solve the problem in dynamic image.Vehicles occluding each other within the image plane especially have caused problem with themeasurement accuracy.In this paper, moving object tracking is investigated and specialattention is paid to occlusion handling, it also become a research focus in the world.This paper focuses on vehicle detection, vehicle occlusion detection and vehicle trackingunder occlusion, and the approaches to solve the problems. The main contributions in thispaper are as follows:(1) For vehicles in the detection of high-definition images, the existing back groundmodeling methods extract the vehicle prospect with too much computation, while exists emptyand chasm in the high-definition images. This paper proposes an dynamic backgroundmodeling and OTSU local recursive segmentation mergetion algorithm. It constructs a mixtureGaussians Model for each pixel, In sequence frames subtracting the model classify the pixelsin each frame into background area and moving objection area. The model set a unique updaterate in uncovered background area. our model moving objection area no longer creates newGaussian distribution, so it can extract the very satisfying vehicle prospect.(2) Occlusion influence the tracking effect. To solve the above issues, we must solve thedetection problems. The paper proposes a pixel statistical model that based on vehicleocclusion detection and segmentation algorithm, which distinguish the blocked vehicle and theblocking vehicle.(3) The vehicle occlusion is easy to cause the loss of tracking. When the vehicle isblocked, the paper proposes a three-frame prediction algorithm based on and then track afterdistinguishing the blocking situation. Through the prediction of motion and occlusionsegmentation, to a certain extent, this paper solve the occlusion problem of vehicle.According to the above algorithm, this paper also designs and performs severalexperiments on the blocked vehicle. In this paper we have accomplished the tracking ofmoving objects in an image sequence using object segmentation framework and featurematching functionality. The experiment results have shown that the proposed algorithm issimple and can solve the occlusion problem to some extent, the accuracy of object tracking isenhanced.