Training device, training method and detection method

Abstract

The invention discloses a training device, a training method and a detection method. The training device comprises a sample acquiring unit, a feature extracting unit and a training unit, wherein the sample acquiring unit is used for acquiring an object image set and a background image set which have preset numbers; the feature extracting unit is used for extracting features of the object image setand the background image set acquired by the sample acquiring unit; the training unit is used for carrying out training by using the features extracted by the feature extracting unit so as to acquirea classifier; the features comprise a wavelet-like feature and a gradient orientation histogram feature; and the classifier adopts a specific cascading structure, combines a Haar-like feature with aHOG feature together for use, plays the advantages of quick calculation of the Haar-like feature and strong resolution of the HOG feature, and simultaneously improves the performance of the acquired classifier greatly without increasing the calculating cost obviously.

Description

technical field [0001] The invention relates to object detection technology, in particular to a training device, a training method and a detection method for determining whether an object exists in a static image or a video image and giving the position of the object in the image or video. Background technique [0002] Techniques for object detection in images and videos use various features to train classifiers in an attempt to find the location and dimensions of objects of various sizes and appearances in images and videos. Taking cars as an example, car detection technology is a branch of the object detection field, and many technologies used for other object detection can also be used for car detection. The problems faced by car detection technology are diverse, for example: cars include various types (cars, trucks, buses, etc.), and there are subtle differences in appearance between cars; occlusion is also a very common problem. The problem of occlusion may occur in va...

AI Technical Summary

Problems solved by technology

[0006] However, there are certain shortcomings in the above three ways

For the first method, since the target is viewed as a whole, the processing speed is relatively slow when testing

The second approach has better adaptability to occlusion than the other two, but the use of multiple local classifiers produces more spurious responses, which introduces new problems

The step-by-step elimination mode of the third way greatly shortens the detection time, so it has been applied more, but its calculation is more complicated

[0011] The advantage of the Haar-like feature is that it is convenient to construct the integral image, the calculation is simple and fast, and it can be trained quickly; when detecting, the method of expanding the detection window step by step instead of shrinking the image obtains a faster speed, but this step by step The way to expand the detection window will slightly reduce the detection rate

On the other hand, the Haar-like feature is a weak feature. In the training of the Cascade classifier, many levels of classifiers are used to reduce the false detection rate, but this also reduces the detection rate of the classifier.

Images