Target detection and identification method and system for real-time video

Abstract

The invention provides a target detection and identification method and system for a real-time video. The target detection and identification method comprises the steps: obtaining the position range CX of a target object in a current video frame of image, wherein the step of obtaining the position range CX comprises: determining whether a target object recognition result RX-1 of a previous frame image of the current video frame is identical to a target object recognition result RX-2 of an image between two frames; if not, performing target object position detection on the image of the currentvideo frame through a first-level neural network, and obtaining a target object position range CX in the image of the current frame image; if so, according to the position range CX-1 of the target object position range in the image of the previous frame, determining the target object position range CX in the image of the current frame ; and carrying out target object recognition on the image of the current frame through a second-level neural network according to the position range CX. Therefore, the operation frequency of the first-stage neural network for position detection can be reduced; the recognition speed is increased; and occupation of CPU and memory resources is reduced.

Description

technical field [0001] The invention relates to the technical fields of artificial intelligence and image recognition, in particular to a method and system for real-time video target detection and recognition. Background technique [0002] With the continuous improvement of recognition accuracy, the method of using convolutional neural network (CNN) for image recognition and classification has been gradually accepted by the industry. When evaluating the performance of models or algorithms, the ImageNet (image network) data set is often used as the test data set. In recent years, various new CNN network structures have been continuously proposed, from the AlexNet (Alex Network) model in 2012 to the VGG (VisualGeometryGroup, Visual Geometry Group) in 2014 and the subsequent GoogLeNet (Google's LeCun network) and ResNet ( Residual Network (residual network) model, CNN has made great progress on the basis of the original LeNet (LeCun network), and the best positioning and class...

AI Technical Summary

Problems solved by technology

In addition, in the application of mobile terminals, the huge heat generated by the high-load operation of the CPU is almost a problem that the hardware of existing mobile terminal equipment cannot solve and bear.

Images