Equipment and method for detecting object based on multiscale comparison characteristic

Abstract

The invention provides equipment and a method for detecting an object based on a multiscale comparison characteristic. The equipment for detecting the object comprises an input unit, a learning machine based on the multiscale comparison characteristic, a sorter loading unit and a detecting unit; the input unit is used for inputting a detected image; the learning machine is used for extracting the multiscale comparison characteristic of a training sample related with the object, and processing the extracted characteristic by using statistical learning; the sorter loading unit is used for constructing a sorter according to a statistical learning processed result of the learning machine based on the multiscale comparison characteristic; and the detecting unit is used for detecting the object of the detected image input by the input unit by using the sorter constructed by the sorter loading unit, wherein the multiscale comparison characteristic is used for instructing comparison relation relative to a specific image attribute among a plurality of relevant rectangular regions under different scales in the image.

Description

technical field [0001] The present invention relates to image feature extraction and object detection technology, in particular, to a device and method for detecting objects in images based on multi-scale contrast features, wherein the multi-scale contrast features are used to indicate multiple Contrastive relationship between two related regions with respect to specific image properties at different scales. Background technique [0002] As the basis of computer vision technology, the detection of specific categories of objects (such as head, face, fist, upper body, shoulder, car, bicycle, package or other objects) is widely used in many fields, for example, human-computer interaction, Action recognition, intelligent monitoring, multimedia retrieval, etc. Due to the academic value of this technology and its broad commercial application prospects, people have paid more and more attention to it, and a large number of researches are especially aimed at real-time object detecti...

AI Technical Summary

Problems solved by technology

However, the problem with the Haar-like feature detection method is that the features themselves are relatively simple. Therefore, in order to achieve high performance, a large number of features must be used to participate in the calculation during detection, which results in an increase in the amount of calculation.

In addition, since the Haar-like feature represents the absolute difference between adjacent regions, this feature cannot resist the influence of ambient brightness and image noise, and cannot obtain effective detection results when applied to images with complex environments

Although people try to obtain better performance by improving Haar-like features or proposing additional features, a satisfactory object detection method has not been obtained in terms of detection speed, calculation amount, and detection effectiveness.

Images