Study On The Vehicle Targets Classification Based On Geometric Features

The Intelligent Transportation Surveillance System (ITSS) aims to extract the interestedobjects in the range of monitoring, and to classify the objects as the ultimate goal. One of thekey issues is the camera calibration. In the paper, we have implemented a series ofexperiments that focus on target points extraction, camera calibration and vehicleclassification. The hardware platform is constructed by a digital camera and a personalcomputer.As the target points extraction is the basis of the camera calibration, it is necessary to studyan efficient, rapid and accurate detection algorithm. This work also directly determines thecalculation accuracy of the camera’s intrinsic and extrinsic parameters. By analyzed theadvantages and disadvantages of the Harris corner detection algorithm and least-squaresfitting curve equation, we propose a new target points detect algorithm base on the imagelinked region centroids. We can extract points precisely through the unicom region by appliedthe image binarization, region tagging, morphological transformation, measurements andother necessary operations. And the algorithm can also be applied to different shapes of thetarget, at the same time avoid the problem that using different targets should use differentdetection algorithm.Camera calibration is a key technology in the current surveillance system. So a simple, fast,efficient calibration algorithm is important. In this paper, we use the classical two-stepmethod and Zhang’s method to calibrate the camera, and verify the principles of the twoalgorithms by experiments. Finally we establish a real-time binocular stereo vision system.Automatically vehicle classification is the core of the research on the intelligencetransportation system. It aims to achieve the vehicle targets accurately classification. Byanalyzed the principles of the swarm intelligence computing, an improved k-means algorithmcombined with cat swarm optimization algorithm (CSO) is applied to vehicle classification.The experimental results demonstrate that the proposed algorithm can classify the vehicleseffectively.