Key Technology Of Object Searching In Video

The key technology of object searching in video mainly contains object detection and recognition. Object detection is the basis of recognition, which has direct influence to the accuracy and robustness of object recognition. There are various object detection methods, different methods are adopted according to the property of objects, such as feature classification, Hu Moments and feature matching algorithm. In this paper, the application of feature matching algorithm in object detection in video is discussed. The following are the main contents of the paper:Firstly, several popular feature points detection algorithms are discussed, including SIFT feature points detection method and the corner detection methods like Forstner, Harris and SUSAN. They are compared by a number of experiments that the invariance to scale, rotation and illumination and the anti-noise ability to Gaussian, and SIFT is determined to be applied in the paper.Secondly, an algorithm of mismatches eliminated based on second-order Gaussian is developed by researching the existing mismatches eliminated methods and SIFT matched points. It is applied in the paper to improve the accuracy of the results of SIFT. In the algorithm, the Euclidean distance between matched points is weighted by a Gaussian function. By comparing these weights, the point whose weight is not in the scale of a threshold is eliminated.Thirdly, the technology of frame extracting is discussed, and DirectShow and FFMPEG is combined to play videos of popular formats, like avi, meg, mpeg, asf, wmv, rm, rmvb, mov, m4v, flv and so on. In addition, a scene detection algorithm based on gray level is developed and integrated into the process of frame extraction to avoid the repeated detection of the frames with the same or similar background and improve the efficiency of object detection.Finally, SIFT algorithm, mismatches eliminated methods based on second-order Gaussian, frame extraction technology and scene detection algorithm based on gray level are combined to implement object detection in video.The experimental results show that the correct rate of object searching in video is higher significantly after mismatches eliminated by the second-order Gaussian algorithm and the efficiency of object searching is improved after the scene detection algorithm is integrated. The object detection method proposed in the paper is invariant to scale, rotation and illumination, and it has a good robustness in object searching in video.