Target detection method and device

Abstract

The invention discloses a target detection method and device. The method comprises: acquiring an initial image; inputting the initial image into a pre-trained deep convolutional network model to obtain a second image which is output by the deep convolutional network model and marked with an interest target candidate box, the interest target candidate box possibly having an interest target contained in the initial image; wherein the deep convolutional network model is obtained by training in advance by using a training image marked with real information of an interest target; wherein the inputof the deepest convolution layer in the deep convolution network model comprises an output result of at least one other convolution layer and a convolution result obtained by performing convolution operation on an output result of an adjacent convolution layer of the deepest convolution layer and network parameters. According to the method and the device, the position accuracy of the candidate boxof the interest target contained in the output initial image is improved.

Description

technical field [0001] The present application relates to the field of target detection, and more specifically, to a target detection method and device. Background technique [0002] Target detection is an important topic in the image field, mainly to detect the possible location and category of target objects in the image. At present, target detection is applied in various scenarios, such as in the traffic field, it is often applied to detect road traffic signs in images, that is, road traffic signs are target objects. [0003] In the field of target detection, image detection technologies represented by deep convolutional networks are often used for target detection. Among them, the deep convolutional network is a multi-layer serial network structure, and the local position sensitivity of the deeper convolutional layer is weaker (that is, the detection ability of the possible position of the target object is weaker), and the semantic sensitivity is stronger ( That is, th...

AI Technical Summary

Problems solved by technology

Among them, the deep convolutional network is a multi-layer serial network structure, and the local position sensitivity of the deeper convolutional layer is weaker (that is, the detection ability of the possible position of the target object is weaker), and the semantic sensitivity is stronger ( That is, the stronger the ability to identify whether the area on the picture is the target object), this feature leads to the greater the deviation between the position of the target object detected by the deeper convolutional layer and the actual position of the target object, which in turn leads to the deepest level. The accuracy of the location of the target object output by the convolutional layer is relatively low

Images