Publication

Abstract : Smart spaces are specialized environments developed to enable the automatic monitoring of events in a monitored setting. Smart surveillance uses deep learning for object detection, to detect any hazards or predict potential threats in the designated smart space. Deep learning improves the accuracy of the dataset and even humans in tasks like image classification, speech recognition, and predictive tasks. In smart spaces, deep learning can be used for actions like voice recognition, to identify trends in collected data and smart surveillance. Deep learning algorithms are capable of locating a region of interest in a frame and predicting a label for the object in the region of interest. There are a wide variety of architectures available, each with its advantages and limitations. This paper aims to provide a study of deep learning architecture performance tuning. After an extensive comparison, considering the given evaluation metrics and time constraints of a real-time smart surveillance system, the YOLO architecture and its variants are found to be the most efficient. This architecture has been implemented on a smart space dataset and the results have been documented.