Research On The Detection And Tracking Technology Of Moving Objects

Moving target detection and tracking is one of the most important issues in imageprocessing and computer vision，it has been widely used in medical detecting,militarystrike, motion coding, intelligent surveillance and other fields. With the continuousinnovation of this algorithm, the practical value of objects detection and tracking is moreand more high.As one of the earliest applications using this technology in real life，however,intelligent transportation system’（sITS）long-term development was limited by limitationsof the detection and tracking technology. The traffic monitoring system needs to work24hours constantly, but most of the traditional detection and tracking algorithms are not ableto reach the real-time and accuracy requirement. Low light condition also makesmonitoring system could hardly work at night. We do much work on the vehicle detectionand tracking to solve the problems in practical application in ITS.In the respect of targets tracking, firstly, a few traditional detection algorithms arestudied, then this paper proposed a vehicle detection algorithm based on improvedAdaboost algorithm and frame differencing rule. To train a cascaded classifier usingHaar-like features of the vehicle samples, while a binary mask image is created by thethree-frame differencing rule, then a foreground mask image was rebuilt by removinginterference noise and finding connected domains, finally the cascaded classifier is loadedto make a multi-scale detection in the current frame’s foreground region. At the sametime, gamma correction is used to ensure the detection rate if the scene is under low lightintensity and the features of vehicle are hard to extract. This method can effectively reduce the detection time of each frame and the false detection rate. It has a goodinhibitory effect on shadow and other interference; it can partly solve the vehicle adhesionproblems when the vehicle occlusion is not too serious. It can provide more time andmore accurate data for vehicle tracking.In the respect of targets tracking,The theories of Kalman filter tracking, particle filtertracking and Mean Shift tracking have been introduced firstly， Secondly an improvedHOG features for vehicle tracking is proposed，it can reduce computation cost andincrease the characteristics of specific target description Then an algorithm combiningKalman filter and Mean Shift algorithm is used to track moving vehicle. on the one handsolves the problem of Mean Shift algorithm can only uses a fixed size window in trackingmoving targets, on the other hand, it overcomes weakness of Kalman filter such as theindependence on model, uncertainty of tracking and Robustness on tracking. By this way,the accuracy, speed and the scale of the tracking system are highly improved.The vehicle detection and tracking system in this paper can be used for intelligentcontrol of traffic intersection and providing more convenient and safe transportationservice to society.