"Port Wide View Vehicle Inspection"

"Wide-field visual perception has become a fundamental component of intelligent port management, traffic monitoring, and digital-twin construction. Nevertheless, the lack of publicly available large-scale datasets tailored to port-level wide-angle scenes significantly constrains research progress in long-range vehicle detection and small-object analysis. To address this gap, this paper presents a dedicated wide-view port vehicle dataset designed for complex maritime surveillance environments. The dataset is collected from multiple real-world port locations using ultra-high-resolution cameras with extensive fields of view, covering diverse operational scenarios such as container yards, shipping lanes, logistics corridors, and mixed traffic zones. It contains a large number of annotated vehicle instances with substantial scale variation, frequent occlusion, and dense spatial distribution. Detailed annotations include bounding boxes, category labels, and visibility attributes, enabling comprehensive evaluation of detection performance under challenging conditions. In addition, the dataset incorporates variations in weather, illumination, and temporal conditions, thereby reflecting realistic operational diversity. Statistical analyses demonstrate significant long-tail distributions and extreme small-object proportions, which better approximate real port surveillance requirements than existing urban traffic datasets. Owing to its scale, annotation quality, and scene diversity, the proposed dataset provides a valuable benchmark resource for research on wide-field perception, small-object detection, and intelligent transportation systems in maritime and industrial environments."