A detection method for small and micro targets in high-altitude video surveillance

Abstract

The present invention discloses a detection method of small and miniature targets in high-altitude video monitoring. The detection method comprises the following steps of (1) aiming at the small and miniature targets, adjusting and optimizing the appropriate candidate area generation parameters when a convolutional neural network is trained, and determining the image length-width ratio parametersof the specified small and miniature target; (2) determining the reasonable minimum detection pixel area of the specified small and miniature target, and realizing a machine intelligent monitoring on-duty function by the timing and positioning 360-degree automatic inspection analysis of a camera. According to the present invention, in order to solve the specified demands and the problems in the high-altitude video monitoring, an object at the algorithm training is adjusted to iteratively approach the generation size and the generation proportion of a candidate frame, so that the algorithm approaching candidate area size fits the actual object better, and a training result of the convolutional neural network and the detection accuracy of the small and miniature object are improved. During aprocess of identifying the algorithm application, the 360-degree automatic inspection analysis is realized by a dispatching mechanism, so that the situation of using the machine analysis and on-dutyto substitute for the artificial video monitoring becomes reality.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to a method for detecting small and micro targets in high-altitude video surveillance. Background technique [0002] With the rapid development of scientific and technological means, video surveillance technology is widely used in various industries. The development of video surveillance has gone through the stage of "visible" (black and white surveillance) to "clear view" (high-definition surveillance), and the next monitoring goal should be to develop to the stage of "understandable" (intelligent image analysis). [0003] In the past, video surveillance systems mostly relied on video equipment manufacturers to build themselves, such as erecting 3-5 meter poles, deploying video servers, decoders, network equipment and storage equipment, etc. to achieve. In recent years, with the rise of cloud computing technology, major operators rely on their own cloud computing center ...

AI Technical Summary

Problems solved by technology

Convolutional neural networks are more effective for large and medium target detection; however, when applied to small and micro target detection, the effect is often greatly reduced

Images