A Machine Learning Pose Detection Framework to Identify Suspicious Activity

Agrawal, Rajat Deepak (2023) A Machine Learning Pose Detection Framework to Identify Suspicious Activity. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration manual) Download (1MB) | Preview

Abstract

Suspicious activity is a type of behaviour that can be classified as unusual and may indicate illicit intent. Detecting these types of activities using conventional methods is a big-time challenge as these methods suffer from computational complexity which limits their real time applicability. This research proposes a machine learning pose detection framework for identifying suspicious activity. Suspicious activities involve unusual movements of the body which might be an indication of a potential threat. By combining pose detection model and classification algorithms this framework tries to enhance the surveillance systems in real time. The proposed framework combines a pose estimation model, an activity classification model, an activity recognition model, and an alert mechanism. The framework proposes to use MediaPipe BlazePose pose estimation model to extract body posture features in the form of x, y, and z coordinates. These coordinates are later used to create customised geometric features and train the model. The framework proposes to create two separate models where the primary model is used to classify if the activity is suspicious or not and the second model is used to recognize the activity if predicted as suspicious by the first model. A large-scale dataset of 1900 real world surveillance videos which consists of 13 different classes is used for this study. Once the features were extracted from this data four algorithms XgBoost, Random Forest, LightGBM, and Deep Neural Network (DNN) were evaluated for both primary and secondary models. Random Forest was determined to be the optimal primary model with an accuracy of 93% and XgBoost was found to be more feasible as a secondary model with an accuracy of 70% in predicting activities. The potential users who can be benefited from this framework include security companies, private businesses, public safety organizations, etc.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Stynes, Paul UNSPECIFIED
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science Q Science > Q Science (General) > Research > Research--Equipment and Supplies > Scientific apparatus and instruments > Physical instruments > Detectors > Remote sensing > Electronic surveillance Z Bibliography. Library Science. Information Resources > ZA Information resources > Research > Research--Equipment and Supplies > Scientific apparatus and instruments > Physical instruments > Detectors > Remote sensing > Electronic surveillance Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Tamara Malone
Date Deposited:	07 Nov 2024 16:56
Last Modified:	07 Nov 2024 16:56
URI:	https://norma.ncirl.ie/id/eprint/7167

Actions (login required)

View Item