Study of Traditional, Artificial Intelligence and Machine Learning Based Approaches for Moving Object Detection

摘要

The automated detection and tracking of objects in motion from visual data like images and videos is a classic problem in computer vision that has gained renewed interest and seen great progress with modern Artificial Intelligence (AI) and Machine Learning (ML) techniques. In particular, deep neural networks have shown unmatched capabilities for visual recognition tasks like classifying, localizing and segmenting objects of interest in complex scenes. A key advancement that has enabled the success of these techniques is the development of convolutional neural network (CNN) for moving object detection (MOD). Various architectures based on CNN can identify multiple objects in an image and draw bounding boxes around them. These networks can be trained on large annotated datasets to learn rich visual features automatically, eliminating the need for hand-coded feature extraction. Pre-trained networks can also be fine-tuned for specific use cases involving detection of relevant objects like cars, humans and animals. This object detection capability provides the foundation for tracking objects across multiple frames in a video. AI, ML, and deep learning (DL) consistently outperform classical computer vision techniques for tasks like simultaneous object detection, semantic segmentation, anomaly detection in videos and predicting object trajectories over time. With innovations in neural network architectures, unsupervised and semi-supervised learning, and leveraging motion and scene understanding beyond just appearances, the performance on these tasks continues to improve. These techniques provide unmatched capabilities compared to traditional hand-coded approaches. With rapid progress in AI, ML and DL, automated and robust motion intelligence has applicability across domains like autonomous vehicles, surveillance, human-computer interaction, augmented reality, robotics, and more. These make the future promising for automated motion intelligence using AI, ML and DL. This paper provides a comprehensive overview of the state-of-the-art techniques and algorithms that leverage traditional, AI, ML and DL for moving object detection (MOD).