TSQ-2024

The dataset consists of 120 LiDAR measurements of truck passes. The data are presented as 2D images and are provided in the 16-bit PNG format. Some trucks were empty, others were loaded with varying amounts of sand.

Image rows correspond to consequent LiDAR scans. The LiDAR scanning frequency is 50 Hz, so two adjacent image rows were recorded with a time difference of 20 ms. Image columns correspond to the LiDAR scanning angle. The data were recorded within the angle range from -30 to 30 degrees relative to the vertically downward direction. The step of changing the deflection angle is 0.25 degrees, the horizontal image size is 240 pixels, i.e. (|-30| + 30) / 0.25 = 240 steps of changing the scanning angle. The pixel values of the image correspond to the measured distance from the LiDAR sensor to the scene and is given in millimeters

The entire dataset is splitted into three groups of 40 images: clean data, which do not contain any significant clutter; data with medium clutter, which contain light snow or small clouds of sand dust; heavy case data with intense clutter, i.e. with heavy snow and large clouds of dust obscuring the truck body.

Each image has a corresponding ground-truth annotation that is given as a JSON file. This file contains a sequence of coordinates of the polygon points in pixel coordinates that correspond to the expertly segmented boundaries of the dump truck body (they are given in the JSON file under "data" key).