Video Activity Recognition for Surveillance Systems
Karani Kardaş · 2020
This study presents the use and comparison of various machine learning methods in recognition of activities that are useful for inferring complex events in surveillance videos. In this respect, the study aimed to identify four basic activities for surveillance videos which are Run, Stand, Walk and Instant Move. Activity learning success was tested using Logistic Regression, Decision Trees Regression, Random Forest Classifier, Naive Bayes Classifier, SVM, Linear SVM, KNN and MLP machine learning methods. The methods were applied to different types of activities consisting of the BEHAVE dataset commonly used in surveillance video event recognition literature. The study shows the comparative performance of the machine learning methods used. Experimental results show that classification procedures have been achieved successfully. Although the percentages of performance obtained by the methods are approximately close to each other, SVM method yields the best activity recognition result.