Counting of Density Crowd Using Multi-Segment Analysis

In machine learning and computer vision, crowd counting is a difficult task. Most earlier approaches concentrated on crowds with a uniform density, that is, a dense or sparse crowd, which meant they were good at global estimate but not so good at local accuracy. We offer a new approach called Multi-Segment Analysis for counting crowd that tries to people count in varied density crowds so improve crowd counting very relevant in the actual world. We present the Multi-Segment Analysis, which is made up of the following key elements. First, the Density Sensitive Segment (DSS) is made up of various subnetworks that have been pre-trained on scenarios with varying densities. The space (local and global)-density information can be captured using this segment. Second, Feature Improvement Segment (FIS) successfully captures local and global contextual information and assigns the weight to each specific feature of density. Finally, the Fusion Segment (FS) integrates local context as well as fuses these specific features of density. Further, Subnetwork RMSE and Subnetwork MAE are also presented as metrics for better analyse performance on local and global predictions. Counting of Density Crowd using Multi-Segment Evaluation achieves advanced recognition accuracy and excellent robustness with space (local and global)-density crowd counting, according to extensive trials on following counting crowd datasets, UCF_CC_50, ShanghaiTech, UCF_QNR and UCSD.