船只目标图像识别研究场景船只图片数据

该数据为采集与加工数据，通过采集近岸摄像头视频，切分出可能含有船只目标的图片帧，然后标注得到船只图片数据。数据可用于船只目标检测、图像识别研究等场景。
结构规模：（1）数据结构：序号、图片名、图像、标注；（2）数据记录条数：200；（3）存证数据为单次数据，为公司自建。1.对采集得到的近岸摄像头视频进行动目标检测，切分出可能含有船只目标的关键图片帧。具体做法是：设定摄像头的扫描模式，分时间点对不同角度进行一定时间的凝视；将记录下的整段视频按摄像头扫描设定的时间点进行切分，得到针对某一角度的固定视角画面片段；对该画面片段进行动目标检测，如果前后两帧图片之间存在较大变动，认为存在动目标，在存在动目标的帧序列中进行采样得到采样图片。2.通过人工点击和半自动标注工具得到船只位置信息。具体做法是：由人工点击采样图片中船舶所在区域的中心点位，由训练好的标注工具对采样图片进行分析，结合人工给出的船舶区域中心点位信息，给出船舶所在区域的包围框信息。3.将采样图片和包围框信息进行存储，其中采样图片进行base64编码存储，包围框信息则以相对于图片“左-上-宽-高”的相对位置数据存储。

This is a collected and processed dataset. It is obtained by capturing nearshore camera videos, clipping image frames that may contain ship targets, and then performing annotations to acquire ship image data. This dataset can be used for scenarios such as ship object detection and image recognition research.

Data structure and scale:
(1) Data structure: serial number, image name, image, annotation;
(2) Number of data records: 200;
(3) The archived data is a one-time dataset, self-built by the company.

1. Perform moving object detection on the collected nearshore camera videos, and clip key image frames that may contain ship targets. The specific implementation is as follows: Set the scanning mode of the camera, which conducts fixed-duration fixed-point gazing at different angles at specified time points; Split the entire recorded video according to the time points set by the camera scanning to obtain fixed-view image segments for a specific angle; Conduct moving object detection on these image segments: if there is a significant change between two consecutive frames, a moving object is considered to exist, and sample images are obtained by sampling from the frame sequence with moving objects.
2. Obtain ship position information through manual clicking and semi-automatic annotation tools. The specific steps are: Manually click the center point of the ship's area in the sampled image; the trained annotation tool analyzes the sampled image, and combines the manually provided center point information of the ship area to generate the bounding box information of the ship's area.
3. Store the sampled images and bounding box information: the sampled images are stored in base64 encoding format, and the bounding box information is stored as relative position data based on the "left-top-width-height" specification of the image.

搜集汇总

数据集介绍

特点

该数据集包含201条船只图片数据，适用于船只目标检测和图像识别研究。数据通过采集近岸摄像头视频并切分出可能含有船只目标的图片帧，然后进行标注，图像以base64编码存储，标注信息包括船只位置的包围框信息。