| Introduction:computer vision technology is inseparable from the rapid development of object detection and recognition technology. The current object detection and recognition has been completely integrated together, so that detection becomes the basis for identification. The type and properties of the object corresponds to a good variety of different kinds of detection and identification algorithms, such as sift, adaboost, Hu Moments and so on. This paper studies video detection and identification technique. How to improve the accuracy of video detection algorithms and the efficiency of training is the main problem.This article mainly includes the following:(A) The Sub-Frame function in the video program combines ffmpeg decoding technology, the new procedure, the frame of the video format can be any format.the addition of a variety of sub-frame are added to the program (such as direct capture, play capture, scene capture, etc,making the video sub-frame more various and powerful.Recognition from the object in the video file, extract the video file first frame (frame), then from the frame to identify all the information we need. Video framing needs the most common video format that can decode the video and extract frames, and eventually as these frames are detected as the image to detect whether there are suspicious users to find the object. The new procedure, the video frame format can be any format, new program support, including avi, meg, mpeg, asf, mov, wmv, rm, rmvb, flv, etc. almost all video formats, including the sub-frame decoding. The traditional framing process involves several commonly used formats such as avi, rmvb and so on.Support the diversification of the video sub-frames:if the direct capture, play capture, scene capture, etc.(B) object detection algorithm in the video scene recognition was added so that the new method of detection than the traditional detection algorithm to improve speed and efficiency, Improved detection algorithm greatly improved the speed and efficiency, processing time, the new algorithm also changed a lot faster. These are the new algorithm is integrated into the scene change detection function. Add a scene change detection, then, the program can ignore the background of a row with the same frame, while the default information in these frames are the first frame has the same information. At this time the efficiency will greatly improve the detection and identification.(C) object detection algorithm in the video will adaboost algorithm and sift, Hu Moments, Algorithm together. Magnitude and angle for the conventional detection algorithm, the new algorithm to detect objects on the diversity of more adaptive, Compared with traditional algorithm, the new algorithm object zoom in, cover up (that is, some other object detection object is covered), rotation, compression, distortion effects are better detected. |
