Edge data management with symbolic representation

Abstract

Nowadays, Internet of Things (IoT) devices are used in various fields such as health-care, agriculture and smart cities. They continuously generate a large amount ofdata needed to be stored and analyzed. Preserving such data, especially image data,presents a challenge as storage resources are mostly limited. Techniques like real-time object detection have been applied to save only some information about theimage data. However, such techniques limit the possibilities of doing a broaderanalysis of image data. This problem underlines the importance of data compres-sion at the edge, the process in which the size of the data is reduced. It has a directinfluence on increasing network bandwidth and decreasing transmission latency.In this work, we present an online-lossy compression algorithm by means of asymbolic representation of image data. Lossy compression, unlike lossless compres-sion, allows a loss of original information to an allowable extent. Thus, only anapproximation of an image can be reconstructed back.The proposed method uses a modified Adaptive Brownian Bridge Aggregation(fABBA) algorithm to compress the selected Urban Tracker image data at the edge.We showed the importance and upper-lower bounds of hyper-parameters suitablefor image data and fixed the hash-map for online compression. Further, as for theanalysis part, we have compared the object detection from reconstructed images andoriginal (not compressed) images.Our results show that the compression algorithm achieves up to 30% reductionand has 0.886 mAP from object detection. Moreover, we showed that dependingon the image processing task, two adaptive hyper-parameters can have upper andlower bounds.