Research And Application Of Data Placement Strategy In Surveillance Video Cloud Computing Platform

With the society's emphasis on security, the number of surveillance cameras in all corners of the city is growing. This results in the massive monitoring video data. Distributed large scale surveillance video processing becomes a trend. The nodes within the cluster may become load imbalance when we use the traditional distributed video off-line processing systems to process the large-scale video data. At the same time,there is no cooperation between the data distribution and the computing in the traditional distributed data processing system, and this leads to more data transmission cost in the cluster. All the reasons above will reduce the computational efficiency of video off-line processing systems.The problem of the traditional distributed video processing system shows that the collaboration between the distributed storage strategy and the distributed computing is very important. First of all, this paper presents a Computing Time Prediction Model (CTPM) for monitoring video data blocks based on the analysis of the current mainstream intelligent video processing algorithms and distributed computing framework. The prediction model utilizes the video data block's feature,such as resolution, frame rate, duration, and the computing power of the nodes in the cluster, and it can predict the processing time required by the video processing tasks at different nodes. Secondly, this paper proposes an Initial Data Placement Algorithm (IDPA) for video data blocks based on CTPM, which uses the Data Block Group (DBG) of video data blocks belonging to a processing task as the basic placement unit, and predicts each DBG's processing time by CTPM, and then places these video data blocks in the cluster by IDPA. Considering that the computing power of the nodes in the cluster will change dynamically which results in a greater error of the IDPA algorithm. Therefore, this paper proposes a Data Rebalance Algorithm (DRA) based on the CTPM. For some nodes with significant changes in computing power, DRA can ensure the cluster load balancing and improve the cluster resource utilization. Finally, in order to verify the video data placement strategy proposed in this paper, an off-line video synopsis algor:ithm based on Spark framework is implemented. Traffic monitoring video is used as the data source.Experimental results show that IDPA and DRA algorithms can significantly reduce the completion time of the distributed surveillance video processing tasks, so that all nodes in the cluster are load-balanced and then the utilization of the cluster has improved.