Realtime Trajectory Similarity Measurement in Intelligent Surveillance Management Systems Islam Md Mahrab

다국어 초록 (Multilingual Abstract)

Intelligent surveillance management systems increasingly rely on the analysis of movement patterns to ensure safety, security, and efficiency in public and private spaces. Among the key tasks is the real time measurement of trajectory similarity, which enables the detection of abnormal behaviors, identification of suspicious activities, and monitoring of individuals or vehicles across large areas. Traditional approaches to trajectory similarity, such as Dynamic Time Warping（DTW）, Hausdorff distance, and Longest Common Subsequence （LCSS）, have proven effective but often face limitations in real time applications due to high computational costs and sensitivity to noise. Recently, machine learning and deep learning models, including Long Short Term Memory （LSTM） networks and Recurrent Neural Networks （RNNs）, have been explored to overcome these challenges by learning trajectory dynamics and improving robustness in noisy and dynamic environments. This thesis proposes an integrated framework for real time trajectory similarity measurement that combines classical distance based algorithms with learning based approaches. The framework is designed to balance accuracy, scalability, and computational efficiency, making it suitable for deployment in intelligent surveillance management systems. Experiments are conducted on simulated and real world trajectory datasets to evaluate accuracy, robustness, and processing time. Results demonstrate that the proposed framework improves real time detection capabilities while maintaining low latency, paving the way for more adaptive and reliable surveillance solutions.

목차 (Table of Contents)

• Introduction 1
• 1.1 Motivation 2
• 1.2 Contributions 3
• Related Work 4
• 2.1 Classical Distance-Based Methods 4
• 2.2 Sequence-Based Alignment Methods 5
• 2.3 Learning-Based Trajectory Similarity 6
• 2.4 Trajectory Analysis in Intelligent Surveillance 8
• Problem Statement 9
• 3.1 Research Challenges 9
• 3.2 Problem Formulation 11
• Proposed Solution 12
• 4.1 Dynamic Time Warping 13
• 4.2 Hausdorff Distance 16
• 4.3 Point-wise Euclidean Distance 18
• 4.4 Longest Common Subsequence 20
• 4.5 Reinforcement Learning 23
• 4.6 Data Preprocessing 26
• Performance Evaluation 29
• 5.1 Dataset Description 29
• 5.1.1 Given Trajectory Analysis 30
• 5.1.2 Monitored Trajectory Analysis 32
• 5.2 Simulation Setup 34
• 5.2.1 Environment Configuration 35
• 5.2.2 Dataset Preprocessing Pipeline 36
• 5.2.3 GPX Preprocessing Pipeline for Messy Surveillance Data 36
• 5.2.4 Experimental Framework 40
• 5.3 Comparison Table 40
• 5.4 2D Trajectory Similarity Analysis 41
• 5.4.1 Dynamic Time Wraping 41
• 5.4.2 Hausdorff Distance 42
• 5.4.3 Point wise Euclidean Error 43
• 5.4.4 Longest Common Subsequence 44
• 5.4.5 Reinforcement Learning 44
• 5.5 3D Trajectory Similarity Analysis 45
• 5.5.1 Dynamic Time Wraping 45
• 5.5.2 Hausdorff Distance 46
• 5.5.3 Point Wise Euclidean Error 47
• 5.5.4 Longest Common subsequences 48
• 5.5.5 RL-Based Algorithm 49
• Trajectory Similarity Analysis Platform 50
• 6.1 System Architecture 51
• 6.1.1 GPX Parse 52
• 6.1.2 Trajectory Illustration 52
• 6.2 Data Quality Check and Input Validation 53
• 6.3 Upload GPX Trajectory Files 54
• 6.4 Algorithm Comparison Table 55
• 6.5 Input GPX File Visualisation 56
• 6.6 Trajectory Visualisation 57
• Drawbacks and Future Work 59
• 7.1 Drawbacks 59
• 7.2 Future Work 60
• Conclusion 61
• Bibliography 62