简体繁体 English

使用HOG功能进行多人检测

[英]multiple human detection using HOG features

原文 2015-05-29 08:58:33 1 1 matlab/ machine-learning/ computer-vision/ svm

I am doing a project on multiple human detection in static images using HOG features with SVM (matlab) in matlab. 我正在使用在Matlab中使用SVM（matlab）的HOG功能在静态图像中进行多个人体检测的项目。 I have taken 20 single human cropped images for positive training. 我已经拍摄了20张人体裁剪的图像进行正面训练。 Do i have to take multiple humans in image as training in order to detect multiple persons? 为了检测多个人，我是否需要接受多人图像训练？ now my software detecting only one human even there are 5 persons in the image. 现在，即使图像中有5个人，我的软件也只能检测到一个人。 Please help. 请帮忙。 Thank you. 谢谢。

1 个解决方案

You need to train your human detector with single humans. 您需要与单人一起训练人体探测器。
When you start your detector you have to make sure that it continues searching even when it found the first match. 启动检测器时，即使发现第一个匹配项，也必须确保其继续搜索。

A first, very basic (and error-prone) approach would be to subdivide the image into multiple overlapping tiles and running the detector on each of those. 第一种非常基本（且容易出错）的方法是将图像细分为多个重叠的图块，并在每个图块上运行检测器。 This way you could check if the detector is able to recognize the different types of humans. 这样，您可以检查探测器是否能够识别不同类型的人。 If it does not, you have found the bug: the detector itself is insufficiently trained. 如果不是，则说明您发现了错误：检测器本身训练不足。 If it does, you just have to force it to keep looking even after the first hit. 如果确实如此，您只需要强迫它即使在第一次命中后仍保持外观即可。