A Fast Detection and Tracking Method of Optical Small Targets on Water Surface

Abstract

The invention belongs to the technical field of computer vision, discloses a fast detection and tracking method for small optical targets on the water surface, and solves the problems of poor real-time performance, low sensitivity to small targets, and inaccurate tracking of bandwidth-adaptive targets in current deep learning-based detection algorithms , including the following steps: step (1): fast detection of small water surface targets; step (2): generation of key frame target models; step (3): fast tracking of target with anisotropic bandwidth adaptive between key frames. The invention realizes autonomous operation without human intervention; only positive samples are included in the training, which reduces the background false detection rate; the network structure is simple and the detection speed is fast; the detection accuracy of small targets is improved; the number of iterations is reduced, and the accuracy and tracking speed.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a fast detection and tracking method for small optical targets on water surfaces. Background technique [0002] Computer vision uses a variety of sensors instead of visual senses as input, and the computer acts as an analog processing input to the brain. With the rapid development of computer computing power, humans rely more and more on analyzing video content through computer vision to obtain useful information in videos. Among them, target detection and tracking is an extremely important part of computer vision. At present, areas such as port security defense, unmanned vehicle obstacle detection, and maritime traffic monitoring all need computer vision to complete the detection and tracking of surface targets. When the illumination changes greatly and there are multiple water targets in the field of view, how to accurately and quickly detect and track mul...

AI Technical Summary

Problems solved by technology

[0006] At present, the problems faced by the rapid detection and tracking of small optical targets on the surface of the water are mainly reflected in: when implementing port security defense and other tasks, the camera fixed on the shore is relatively far away from the target, and the captured targets contain many small targets. Combining low-level features and high-level features to detect targets without increasing the amount of calculation, especially small targets are problems faced by small targets on the water surface; the bandwidth updates of the existing kernel function bandwidth adaptive target tracking algorithms are almost isotropic , for the lack of accuracy and stability in the tracking of targets that just appear and slowly disappear or rotate, it is easy to cause errors to accumulate continuously during long-term target tracking, resulting in target drift or even loss

Images