Multi-object Tracking Method Based On Video Stream For Indoor And Outdoor Environment

With the growing popularity of security cameras in public and work-related places,there is an increasing demand for visual object tracking.Currently,object tracking systems are being utilized in several key applications including video surveillance,traffic monitoring,smart environments,etc.In view of the contemporary research related to computer vision,object tracking is one of the challenging and emerging topic in the domain of image processing.In object tracking,we are encountered with numerous constraints that affect the performance of object tracking,such as object-to-object occlusion,object-to-scene occlusion,illumination variations,dynamic background and noise in a frame,etc.In this thesis,we incorporate several approaches in indoor and outdoor environments in order to resolve the forementioned problems.Firstly,we analyze the mean-shift tracking(MST)algorithm because of its robustness and simple design.However,MST has some limitations when it is used in challenging conditions,such as occlusion and variations in illumination parameters.In order to address the partial and full occlusion problems,we propose a method for single object tracking(SOT)based on image similarity measures(ISMs)such as normalized cross-correlation(NCC)and NCC-extended methods.The experiment results on different indoor and outdoor scenarios demonstrate the effectiveness of our proposed approach after utilizing a tradeoff strategy against various tracking parameters.Thus,we achieve a fast and reliable object tracking in occlusion related scenarios.Furthermore,we also propose a multi-object tracking(MOT)method for both indoor and outdoor environments.The proposed method is based on numerous features and attributes of computer vision,such as adaptive background subtraction,running average and image morphology.Additionally,adaptive background subtraction is performed based on the Gaussian mixture model(GMM).We demonstrate the feasibility of our proposed approach by introducing changes in illumination,creating dynamic backgrounds and occlusion related scenarios.Our optimized approach offers cognition ability in the design,thus making the proposed MOT method self-governable and self-configurable.Our simulation results depict that the MOT method can track multiple objects accurately with an accuracy rate of 85.71%.