Population counting method based on scale-adapted head detection and density map

Abstract

The invention discloses a population counting method based on (a) scale-adapted head detection and density map. Characteristic training and prediction are carried out on an image; gradient informationand a foreground are extracted from the image; the scale and parameters corresponding to the image are generated; a foreground image is segmented, samples are screened; a training model of the head is obtained by training the samples; a training model is used to implement prediction, and a prediction result is obtained; and multiple scales of density maps are generated according to the predictionresult, and the density maps are added to obtain the total prediction number. Pedestrians in the image are counted by combining scale-adapted method with head detection, and the disadvantage in perspective transformation of a common detection method is overcome; and the adaptive scale screening method and the density map are used to obtain a higher robustness of the method, and the method can beapplied to different scenes; and a trained model has a higher classification capability via patch screening and classification, and the accuracy of population counting is guaranteed.

Description

technical field [0001] The invention relates to the field of pedestrian detection, and more specifically, to a crowd counting method based on scale-adaptive head detection and density maps. Background technique [0002] With the development of social urbanization and the rapid increase of urban population, video surveillance is increasingly used in daily work and life. One of the most important application areas of these video data is intelligent video surveillance. In China, which has a population of 1.375 billion, a series of problems caused by a large population have always threatened public safety. Also in other parts of the world, when large-scale events are held, uncontrollable events will occur due to crowds. Therefore, the effective use of security monitoring data to rationally deploy security personnel and build auxiliary transportation facilities to guide and divert the crowd is of great significance to the maintenance of public order and the protection of person...

AI Technical Summary

Problems solved by technology

Although the human body recognition and tracking technology is relatively mature, the detection accuracy of the whole human body in complex scenes will be lowered due to problems such as occlusion and perspective transformation.

In addition to traditional methods, deep learning also has excellent performance in crowd counting. The advantage of deep learning methods lies in higher accuracy. Through long-term training and parameter adjustment, deep learning can achieve better accuracy than traditional methods. But the disadvantages come accordingly. Training requires a large amount of training data and takes a long time to train. It takes up a lot of computing resources and requires high-performance computing equipment to cooperate with it. This makes the application more difficult and the equipment cost becomes higher.

Images