AI-driven Learning System: An Automated Moving Object Detection Approach for Complex Surface Scenes
A surveillance system is necessary for monitoring and ensuring safety across diverse natural environments. Core to this system is change detection followed by object tracking, yet detecting moving objects at varying speeds remains difficult. Fast-moving objects often cause motion blur, obscuring critical details, while slow-moving objects can be misclassified as static, particularly in cluttered urban scenes or dense forests. Occlusion and camouflaging objects further complicate tracking in crowded or visually complex settings, and performance frequently degrades in unseen video scenarios. Traditional methods often struggle to address these challenges effectively. This work presents a novel approach to overcome these limitations, emphasizing the importance of background subtraction (BgS) for reliable change detection. A new encoder–decoder architecture is proposed for detecting moving objects in complex video environments. The encoder employs a modified VGG-16 network to capture fine- and coarse-scale features critical for accurate detection, while the feature extraction framework (FEF) integrates average pooling, convolutional layers, and convolutional layers with varying sampling rates to achieve multi-scale, multidimensional feature representation. The decoder, constructed with stacked transposed convolution layers, contrast normalization, and LeakyReLU activations, efficiently reconstructs these features into image space. Extensive evaluations against various existing methods demonstrate superior performance, including urban surveillance and animal detection in forest environments.