Fish silhouette for fish pass monitoring

The dataset is an output of the TAČR project SS07020014 - System for fish migration analysis and detection of invasive species. It was recorded for the purpose of the project to develop a monitoring device for fish pass monitoring. The date was recorded by the near-infrared camera with the near-infrared back illumination. It consists of four ZIP files. Each zip file contains two folders: videos and images. The video folder contains raw video records of the fish passing through the monitoring system. The name of the video file contains the name of the fish species passing through the system. The Images folder contains subfolders. Each subfolder contains a sequence of images. The images correspond to the one fish pass. The data was used for the training of the detection and classification system of the fish species passing the monitoring system. The images represent the silhouette of the fish because the backlight illumination was used for recording. It can be used for research on the shape of the fish and fish detection and classification. The dataset contains 19 fish species.

The data were acquired using a custom underwater monitoring system developed within TAČR project SS07020014. The system consists of a waterproof housing, a monochrome Econ See3CAM_20CUG camera, near-infrared illumination and an NVIDIA Jetson Orin Nano processing unit. The camera was equipped with a 9 mm band-pass filter transmitting wavelengths around 840 nm and a 90° lens, providing an approximate field of view of 106° horizontally and 88° vertically.

Videos were recorded at 180 frames per second with a resolution of 640 × 480 pixels. The extracted images have the same resolution of 640 × 480 pixels. The camera was installed with a 90° rotation to make effective use of its sensor resolution when imaging the narrow passage through the monitoring device.

Fish were recorded using near-infrared back illumination. They therefore appear primarily as dark silhouettes against the illuminated background. Near-infrared illumination enabled recording under both daytime and night-time conditions without exposing the fish to visible light. Illumination intensity and camera exposure time could be adjusted according to water turbidity and local recording conditions.

Each video directory contains a FramePos.txt file defining the position of the illuminated region of interest within the 640 × 480-pixel frame. The file contains four coordinate values specifying the boundaries of the illuminated rectangular region. Pixels outside this region were not back-illuminated; consequently, a fish silhouette is generally not visible there and this part of the frame should not be used for silhouette-based analysis. For the provided example, the illuminated region extends across the full image width (x = 0–640) and between y = 210–300.

The dataset includes original video recordings and corresponding image sequences. Each subdirectory in the images directory represents a single fish passage and contains sequential frames from that event. Video filenames include the corresponding fish-species label. During targeted data acquisition, fish were introduced into the monitored passage separately by species, allowing the taxonomic label to be assigned from the known species present during recording. Multiple individuals and size classes were recorded where available.

The dataset contains recordings of 388 individual fish representing 19 species: bleak (79 individuals), chub (45), perch (35), roach (31), silver bream (30), Wels catfish (21), brown bullhead (19), topmouth gudgeon (17), pike (16), Prussian carp (14), barbel (12), pumpkinseed (12), common carp (11), pikeperch (10), common bream (10), asp (8), burbot (8), European eel (5), and vimba bream (5). The dataset includes both native and non-native species, including invasive species relevant to the project’s objective of automated invasive-fish detection. Each count refers to a recorded individual fish; because every passage is represented by a sequence of frames, the number of images is substantially higher than the number of individuals.

Example of image of perch extracted from the dataset: