Description

In video surveillance applications, the main aim will be identification of the pedestrians i.e., identifying the position of the persons in each frame of the video. In real world, many tracking tasks suffer from the multimodal likelihood and posterior and inaccurate local evidence. The abrupt movements and scaling of the objects in the video always remains a problem for tracking process. The objects can be tracked using many methodologies such as object based tracking and position based tracking. The object based tracking methods identifies the objects in the video using background subtraction and tracks the objects. The object based methods cannot be able to handle occlusions. The position based tracking methods tracks the objects in the video based on the initial position of the target and updating the position of the target by learning of the position of the target in the consequent frames. The input video is converted into frames. The frames were preprocessed to remove the noises from the frames. The objects are detected from the frames by initially identifying the object regions by applying three frame differencing and EMGMM. The objects in the Region of Interest is counted. The SURF feature points were extracted from the input video frames. The irrelevant feature points were removed based on the comparison of detected feature points and segmented objects. The detected objects were tracked based on the optical flow detection based on Lucas kanade optical flow detection process. Finally the performance of the process is measured by measuring the error rate and accuracy. The measure performance of the process shows that the proposed method is capable for the identification of the object positions in a accurate manner compared to the other existing methods used for the identification of the object positions in the video which is due to the intensity based feature values extracted from the image that denotes the object related information in a better manner.