A Crowd Counting Method Generated Based on Training Data of Deformable Gaussian Kernels
Crowd counting,as an important sub topic in computer vision and pattern recognition tasks,plays an extremely important role in intelligent monitoring.For crescent-shaped human heads that are severely occluded,the crescent-shaped visual center found by traditional Gaussian kernel generation methods deviates significantly from the complete circular center annotated by humans,making it difficult for the algorithm to converge during training.A crowd counting method generated based on training data of deformable Gaussian kernel is proposed to address the issue of crowd counting errors in severe occlusion situations.The method adjusts efficiently the shape,angle,and position of the Gaussian kernel generated based on human calibration results,thereby improving the convergence and accuracy of the algorithm.The experimental results show that this method can significantly improve the performance of crowd counting.
国家科技期刊平台
NSTL
万方数据
