Object detection method and device for screening obstacles for visually impaired persons based on positive samples

Abstract

The present application discloses a target detection method for screening obstacles for visually impaired persons based on positive samples. In this method, all candidate frames of each obstacle in a target image are first extracted, and each candidate frame is classified to calculate each candidate frame Then calculate the degree of overlap between each candidate frame in the target image and the marked frame, and then perform weighted fusion of the category confidence and overlap of the candidate frame of the obstacle, and judge the candidate according to the result of the weighted fusion. Whether the box is a positive sample, if so, detect it according to the current target type and positive sample to get the detection result. Because in the technical solution of the present application, the category confidence and overlapping degree of the candidate frame of the obstacle are weighted and fused, so that the category confidence and overlapping degree corresponding to the obtained positive samples are both high, which improves the The accuracy of the positive sample can obtain a more accurate position of the obstacle detected by the target, thereby improving the detection accuracy of the target detection.

Description

technical field [0001] The present application relates to the technical field of positive sample screening, and in particular, to a target detection method and device for screening obstacles for visually impaired persons based on positive samples. Background technique [0002] At present, since many target detections related to visually impaired people involve the screening of positive samples of target obstacles, for example, to predict the location of obstacles, target detection needs to filter out positive samples from pictures, and then detect and locate obstacles based on positive samples. Therefore, it is particularly important to screen out positive samples with higher accuracy. [0003] The existing positive sample selection method is as follows: a candidate frame whose overlap between a candidate frame extracted from an obstacle in the picture and the real label frame of the obstacle is greater than a certain threshold as a positive sample. However, the category co...

AI Technical Summary

Problems solved by technology

However, the category confidence of positive samples with a large degree of overlap is not necessarily high, which leads to non-maximum suppression in the subsequent non-maximum value suppression, positive samples with higher confidence will have a negative impact on other positive samples with lower confidence but high overlap. As a result of misjudgment, some positive samples with low confidence but high overlap will be filtered out, so that the accuracy of the obtained positive samples will be low, which will affect the detection accuracy of the subsequent final target detection, so that the detection results of target detection will not be correct. precise

Images