Research And Implementation Of Population Density Estimation Methods

There are many algorithms of crowd density estimation. But due to complex environment, no algorithm can get satisfactory result in all occasions. Selected a suitable algorithm will improve the detection accuracy. According to different density of crowd there can be divided to low density, high density and super high density. According to qualitative statistical and quantitative statistical, the algorithm can be divided into two broad categories. As a whole, super high density and high density should be counted by qualitative statistical. Low density should be counted by quantitative statistical. As for directional scenes or simple scenes such as one-way street, we can analysis crowd density through calculating the number of individual like face detection or histograms of Oriented Gradients. The method of Crowd density estimation based on pixels count and texture analysis have their own advantages and disadvantages. The key is to select a suitable one according to different scenes.Traditional image processing architecture captures images by front-end and processes images by back-end. This method need large bandwidth to communication and it also need a large consume to central processing unit so that it is difficult to achieve the algorithm. The best solution is distributing more work to front-end, processing the captured images in real time. Only in this way we can reduce the cost of bandwidth. High speed FPGA can process the images concurrent, so it can marked raise the processing ability. Images processing system often be divided two categories according to different levels. Different categories needs different processing system. Low-medium level algorithms have many repeated calculation if using the concurrent nature of FPGA can optimize the performance. In the other side, high-medium level algorithms have many judgements and jumps so it usually processed by central processing unit.The main work of this paper is analyzing different algorithms and improving them; In image processing system many work be arranged to front-end. The mainly innovations of this paper as below:1. Divided the level of algorithm of image processing, arranging low-medium algorithm to front-end and arranging high-medium algorithm to back-end.2. As for images captured by sensor using NNI to recovery RGB, processing speed is fast and the system only take a little resource of FPGA.3. Improving the texture analysis algorithm by adding wavelet and filter, This method effectively improves the accuracy in medium crowd density scenes.4. As for the algorithm based on histograms of Oriented Gradients, the angle of view 45 degree is better than 0 degree.5. Analyzing different algorithm of crowd density estimation, summing the advantages and disadvantages.