Dense Crowd Counting Algorithm Based on Coordinate Attention Mechanism and Inverse Distance Mapping

Aiming at the problem of high density crowd with complex background and poor crowd density estimation ability, a dense crowd counting algorithm based on coordinate attention mechanism and inverse distance mapping is proposed. Firstly, the coordinate attention high-precision network CA-HRNet is created to obtain the head and position feature information of the crowd. Secondly, the focus inverse distance map is obtained by combining the position information of the crowd head and the focus inverse distance mapping to accurately located the center position of each person's head in the crowd. Based on the response difference between background and crowd head in inverse range mapping, the background and crowd head are distinguished to improve the anti-interference ability of the network. Finally, the average maximum filtering strategy is constructed to count the center points of the head. The background noise and similarity points are effectively filtered through this strategy to improve the crowd density estimation ability. The crowd density estimation algorithm was selected to conduct comparative experiments on the public data set. The experimental results show that the proposed algorithm is superior to the compared crowd density estimation algorithm. The proposed algorithm has better ability to suppress background interference and can estimate the crowd density well in different scenes.