Target detection method and device

Abstract

The present invention provides a method and device for target detection. In the method, the codebook is learned based on the initial training sample set, the visual word network model is constructed based on the codebook, and the word frequency histogram features of each sample in the current training sample set are extracted, and the network adjustment operation is performed. Get the weight of each visual word, determine the kernel function of the k-nearest neighbor classifier according to the weight, determine the k-nearest neighbor classifier, detect the area to be detected according to the k-nearest neighbor classifier, and then determine whether the area to be detected is a target. If the detection task is not completed, after obtaining the training samples online, update the training sample set, then update the visual word weights through network operations, adjust the kernel function according to the weights, and then detect the area to be detected. During this process, the k-nearest neighbor classifier can self-adjust at any time according to the current situation, maintaining its adaptability to the current target and background appearance rules, so as to achieve reliable and real-time target detection in changing scenes.

Description

technical field [0001] The invention relates to image detection technology, in particular to a target detection method and device. Background technique [0002] Object detection refers to the technology of identifying and locating objects of interest in images or image sequences. This technology is based on computer vision and is widely used in military, civilian and other fields. [0003] The detection algorithm used in target detection technology is also called a detector. From a structural point of view, the detector includes a feature extraction module and a feature classification module. Among them, the feature extraction module is used to abstractly describe the appearance of the area to be detected as "features" according to certain rules, and the feature classification module is used to classify the area to be detected as a certain type of target or background according to the features extracted by the feature extraction module . In the target detection process, d...

AI Technical Summary

Problems solved by technology

The change of the detection scene makes the originally designed detector not suitable for the target detection in the changed detection scene, resulting in low detection accuracy

Images