Flag detection method

Abstract

The invention belongs to the field of object detection, discloses a flag detection method, and provides a mixed solution, which comprises the following steps: enhancing an original flag data set by using a plurality of effective data enhancement methods; carrying out target detection in the first detection branch by adopting an Optic Flow and GMM combined method; in the second detection branch, taking the video frame input of the expanded data set as the input of a Darknet-53 backbone network to extract the feature map layer of the multi-zoom video frame, using a sample selection algorithm toselect positive and negative samples, and then training a yolov3 deep neural network model and target detection; and combining the detection results of the two detection branches to detect whether a flag exists in the video stream of the camera. The invention has the advantages of high accuracy, low omission ratio, low false detection rate and real-time performance.

Description

technical field [0001] The invention belongs to the field of object detection and relates to a flag detection method. Background technique [0002] At present, the means of illegal information mainly rely on human identification. Facing the increasingly large amount of data, manual inspection is very weak. Therefore, how to use computer technology to realize automatic and intelligent network illegal information retrieval is of great significance. [0003] As a kind of symbol, the flag is widely used in the transmission and expression of information because of its simple and rich expression of meaning, emotion and command actions, and often appears in complex scenes such as parades and rallies. However, due to the non-rigid nature of the flag, the flag cannot be recognized under conditions such as local deformation, partial occlusion, and drastic changes in light intensity. Therefore, there are not many researches on flag detection at home and abroad, but some researchers h...

AI Technical Summary

Problems solved by technology

Ahmed K. (Ahmed K, Rahman M Z, Shameemhossain M.Flag Identification Using Support Vector Machine[J].Juniv Edu,2013) and others proposed a SVM flag recognition system based on the color ratio, which can recognize a maximum of 45 degrees Flag images within corners, but images with complex scenes and multiple targets cannot be recognized

It can be seen that for the research on flag recognition, domestic and foreign researchers mainly use traditional shallow machine learning algorithm models. These models have been proved to have their own limitations and cannot be applied to complex scenarios. Samples require a lot of manual cutting and labeling work

[0004] In addition, due to the soft and light texture of the flag, it is very easy to deform the flag, and the complex environment of the parade and assembly may easily cause the flag to be partially folded and blocked, and it may also easily bring about drastic changes in the intensity of light. In the picture, due to Brightness, low resolution, background, flag shape changes, and influence changes make flag detection face greater challenges

In this case, the existing technology has certain difficulties in identifying flags, and it is prone to missed detection and false detection

Images