Using advanced ML for improving surveillance accuracy

An increase in research over the past 60 years in the field of machine learning widened its areas of application from merely making computers learn to play board games to analysis of big data. Many algorithms have been developed that are now commonly used in various fields ranging from natural language processing to computational finance and has been brought to use commercially as well. Recently, there has been an increase in research on machine learning application in the area of automated video surveillance systems. Most of these algorithms assume that both the training data and test data belong to same feature space with same distribution which might not always be true. This constraint gave rise to the concept of transfer learning which uses the knowledge from the preoccupied knowledge from other related task. This paper aims at improving the efficiency of a transfer learning based machine learning technique for object classification, MKTL framework. It can be brought to use for multiclass object classification in automated video surveillance systems.