AGVS-T22 contains 37 airport videos, nearly 120,000 frames, and ground truth. Each video contains some challenges in airport scene surveillance, such as occlusion, illumination changes, and multi-scale objects. We believe that the challenges mainly come from three sources: 1. The object itself, i.e., aircraft; 2. The environment, i.e., weather; and 3. The specific imaging mode, i.e., infrared image. Below, we will provide a detailed introduction to the challenges faced in the airport.
AGVS-T22 consists of 37 long videos, 120,000 frames and corresponding change detection ground truth. AGVS contains multiple challenges.
Appearence: Aircraft have unique shapes and motion patterns. For example, most airplanes are predominantly white in color and have similar sizes and shapes. This makes it difficult to extract the appearance of airplanes. The shape of an airplane is unique, with a long, slender fuselage and multiple elongated protrusions, such as tail wings and side wings.
Simultaneous multi-scale objects: Due to the vast expanse of the airport area, the images captured often exhibit the presence of multiple objects at different scales. Some aircraft appear at a considerable distance from the camera, rendering them very small or even imperceptible to the naked eye in the image. However, within the same image, other aircraft can be observed with distinct outlines and even visible details. Despite the absolute scale, the majority of objects in the imaging plane are relatively small.Simultaneous multi-scale objects is a key focus in airport, and one of the important challenges lies in ensuring that the detection of small-scale objects does not compromise the detection of objects at other scales.
Motion pattern: In addition to ground-based movements similar to cars, airplanes have unique motion patterns, such as high-acceleration takeoff and landing. Such special motions pose requirements for the position prediction component of algorithms.
Weather: The weather conditions, including haze, rain, and cloudy skies, present numerous challenges to multiobject tracking in outdoor environments such as airports. In haze conditions, due to the vast area of airports and the generally low resolution of targets, the transparency of the air decreases significantly with increasing distance, resulting in severe blurring of video images. During rainy weather, accumulated water on the ground below the flying object can cause serious reflection effects. The striped patterns left on the ground after the rain can also affect the recognition and detection of flying objects. In heavy rain, rainwater can even dripontothecameralens,causingimagedistortion.Incloudy conditions, the airport is segmented into different regions by clouds, and the appearance of flying objects continually changes.
Illumination change: The airport scene exhibits rich variations in lighting conditions, such as gradual changes in illumination during different times of the day, including morning, noon, and evening.
Artificial environment: The airport premises are maintained in a clean and orderly manner; however, airport videos typically capture complex background scenes.
Specific imaging mode: These different imaging modes capture video data with significant variations, posing challenges for adaptation by conventional algorithms.