Analyzing the Performance of Object Detection and Tracking Techniques

Abstract

Objects are detected in computer perspective widely in many real-world applications. In case of video processing, detection and tracking of objects should be very proper and effective. Objects are detected and tracked by traditional methods such as background subtraction, optical flow, and frame differencing method. Convolution neural network, which is a deep learning-based approach, is recently adopted by many developers to identify the object. In this paper, methods of object detection are implemented, analyzed, compared, and discussed. Out of which a robust method has been suggested which satisfies the parameters of precision, recall, and accuracy, also the visualized parameters like object localization, classification, and forms a bounding box to the object are observed and analyzed. It is observed that convolution neural networks detect all relevant objects more accurately than traditional methods. CNN locates the identified object in a video frame using a bounding box that extracts the feature and trains the image for classification. Here, CNN is considered the most promising method for object detection and tracking and can be used in further study where complex work to be handled based on object detection like video inpainting or video restoration.