Ship target real-time detection method based on cradle head camera

Abstract

The invention relates to a ship target real-time detection method based on a cradle head camera. The ship target real-time detection method comprises a step of acquiring a video stream including a ship target, a preprocessing step of training data, a step of determining a training model and improving calculation speed of a training ship target detection model, a step of adjusting the training model according to actual conditions, and a step of linking the trained ship target detection model and the cradle head camera. By adopting the training data preprocessing of the video stream, determiningthe training model, improving the calculation speed of the training ship target detection model, adjusting the training model, and finally linking the trained ship target detection model and the cradle head camera, a width of a ship object in a display main code stream occupies one third of a width of a frame, and is displayed in a middle part, and therefore an object of cradle head camera continuous tracking is realized. The ship target real-time detection method has advantages of scientific and reasonable design, smart concept, easy realization, fast calculation speed, wide popularization and application value, and ability of effectively improving accuracy.

Description

Technical field [0001] The invention belongs to the technical field of video monitoring, in particular to a real-time detection method of ship targets based on a pan-tilt camera. Background technique [0002] In recent years, based on the development of deep learning, in the field of industrial embedded products, due to the enhancement of computing power of computing components, products have achieved rapid development in the two major algorithms of image recognition and speech recognition. [0003] The task of applying artificial intelligence to products to produce smart products is to hope that the product can interpret the data obtained by cameras and other sensors and make decisions in the rational way of thinking that humans have, so as to help humans complete various boring and repetitive decisions. jobs. Most of the existing equipment using traditional image processing methods only recognize moving targets on the sea at a long distance, but there are difficulties in classif...

AI Technical Summary

Problems solved by technology

Most of the existing equipment using traditional image processing methods only recognize moving targets on the long-distance sea surface, but there are difficulties in the classification of moving targets, especially in the detection of large target ships in complex backgrounds after zooming in.

[0004] In recent years, the existing training models are effective for large target detection, but they cannot detect remote small targets, especially boat-shaped objects in hazy weather.

Most of the deep learning method training models with high detection and recognition rates are slow to process a single frame due to the large amount of calculations, so it is difficult to apply to the front-end embedded platform for real-time video stream detection

Can not real-time, can not zoom recognition, can not be applied to the front-end equipment to protect the privacy of user data and other shortcomings, are not suitable for the alarm and tracking requirements of surface ship observation and recognition

Images