Research On Objects Detection And Tracking Algorithm In Warehouse Monitoring Region Based On Machine Vision

Intelligent safety management will be one of the most important technical means ofmodern warehousing safety management, which is an important part of the warehousemanagement. Machine vision can detect and track motions in warehouse surveillancearea through digital video monitoring information extraction and intelligence analysis.When any suspicious behavior in this surveillance area is judged, the system automaticissue a alarming signal, remind related staff who is charging for these regions to dealwith.This paper actualizes a system capable of detection and tracking objects under somegiven warehouse regions, premise on the premise of fixed camera.The main contributions of this thesis are as follows:（1）Moving objects detection: Gaussian Mixture Model was the most popular modelfor imitating background to be subtracted by corresponding image because of adaptability.While it often failure to detect objects because of light mutating or camera shaking, and itneeds large number of computer space. So an improved algorithm of moving objectdetection based on GMM was presented on this paper. We will focus on the robustness byadjusting parameters of the model through judging light mutating and camera shaking.Also, convergence speed was improved by adjusting the means and variance of themodels. We also need preprocessing before detection and morphological filtering whenfinished target extraction due to the noise from image acquisition. A shadow removingalgorithm based on RGB color space. The experiment result shows that this algorithm caneffectively remove the shadow cast by moving object and maintain acceptablecomputational complexity.（2）Moving objects tracking: In the analysis of the current tracking algorithm, We dofurther research on the Kalman filter algorithm to predict and track one moving objectbecause of the reduction of real-time tracking moving object. We choose centroid, widthand height of minimum bounding rectangle of targets as the input parameters of stateequation and observed equation, then match these features around predicted range. On theresearch of multi-target tracking, this paper gives the Kalman filer and tracking model abrief presentation and proposes a method of combining multi-feature fusion and Kalman prediction for multi-target tracking. The center, area and velocity of targets are composedof integration measure, which is taken to match the multiple targets. Experimental resultsshow the good accuracy and stability of using this method to track multiple movingtargets.