Research On Target Detection Algorithm In High Definition Video

With the development and wide application of computer and image process technology, the intelligent crowd density monitoring system has become the focus of research. In this paper, in view of people counting and crowd density analysis in video surveillance these two typical applications, four different algorithms are used to research and compare. From two aspects, the paper puts forward the solutions of these two aspects respectively. Its main contents include:The first aspect is the method of counting human head. Through a lot of research, in view of the low, middle and high these three kinds of different crowd density level, technical route is determined. Some reasonable technical methods are chosen for the feature detection algorithm of crowd density and of the crowd flow, and also tested to determine the two technical solutions for crowd detection algorithms. For the feature detection algorithm of crowd density, foreground image is obtained by carrying out the image segmentation and frame difference of moving people, and the crowd block of the image is obtained by implementing expansion and edge tracking. And the rough number of crowd individuals in the current block is calculated to estimate the crowd density level through the way of the block-within edge detection to search the individual ch aracteristics of the human body. For feature detection algorithm of the crowd flow, the detection of the crowd flow is carried out by setting the virtual doors and specified areas to detect the people within. The basic detection idea is to extract the foreground image through image segmentation and frame difference of moving people. After edge detection, the number of crowd individuals within a demarcated area is detected by the way of searching the individual characteristics and the video is qualified through the detection of the body crossing virtual door realized by continuous multiframe tracking the body motion trails. Then, the software is developed. Test verification software interfaces of feature detection algorithm of crowd density and feature detection algorithm of the crowd flow are respectively built. Finally, the debugged and tested software is employed for the statistics of individual number.The second aspect is the analysis of crowd density, the crowd density is divided into four levels which are sparse, normal, saturated and alarm. The array symbiotic matrix method is applied for the feature extraction of crowd density. A software is developed and the interface of feature extraction which is carried out by array symbiotic matrix method is obtained. The results of the experiment live up to the desired effects. In order to achieve better detection of objects in the case of a high density, a further optimization for feature extraction with a combination of the wavelet transform and array symbiotic matrix method is implemented. Recognition rate is significantly improved. Then the support vector machine is employed to realize the estimation of crowd density level and get the level information of crowd density. Employing wavelet transform and array symbiotic matrix for feature extraction is more suitable for people counting in high density.