Multi-scale feature estimation and high-order BING feature-based target detection method

Abstract

The present invention provides a multi-scale feature estimation and high-order BING feature-based target detection method. The method includes the following steps that: an image feature pyramid is constructed for an original image by using an existing multi-scale histogram of oriented gradient approximation algorithm; high-order BING features are extracted in each histogram of oriented gradient; each window in sliding windows is judged by using a two-stage linear SVM; and whether the windows contain a target is determined. According to the method of the invention, the BING features are expanded to the histograms of oriented gradient from simple gradient features, wherein the histograms of oriented gradient can show more details of the features, high-order statistical information is extracted, and therefore, feature expressions can be richer; and when the feature pyramid is calculated, the multi-scale histogram of oriented gradient approximation algorithm is adopted, and therefore, a more detailed feature pyramid can be obtained under a condition that calculation quantity is not increased, and the position of the detected target is more accurate.

Description

technical field [0001] The invention belongs to the field of pattern recognition and relates to target detection technology. [0002] technical background [0003] As the bottom-level technology of computer vision, target detection technology has always been one of the cutting-edge research directions in the field of computer vision. The upper-level algorithm further analyzes and processes the detection results to realize applications such as target tracking and recognition. In recent years, many new target detection algorithms have been proposed, such as the FPDW algorithm based on the boosting idea, the saliency detection method based on BING features, and the fast-cnn algorithm based on deep learning, but they cannot be achieved in terms of detection accuracy and detection speed. have both. The FPDW algorithm has a high detection accuracy, but the detection speed is only 15fps; the detection speed of the BING algorithm can reach 300fps, but the detection accuracy is very ...

AI Technical Summary

Problems solved by technology

In recent years, many new target detection algorithms have been proposed, such as the FPDW algorithm based on the boosting idea, the saliency detection method based on BING features, and the fast-cnn algorithm based on deep learning, but they cannot be achieved in terms of detection accuracy and detection speed. have both

The FPDW algorithm has a high detection accuracy, but the detection speed is only 15fps; the detection speed of the BING algorithm can reach 300fps, but the detection accuracy is very low, and can only be used as a rough estimate of the target position; the fast-cnn algorithm is a detection based on deep learning Algorithms, high demands on hardware

Images