Video-based pedestrian and crowd behavior identification method

Abstract

The invention discloses a video-based pedestrian and crowd behavior identification method. The whole framework comprises a single-person limb part network, a single-person whole and limb joint networkand a multi-person network. The framework aims to jointly learn joint co-occurrence and time evolution in an end-to-end mode, and joint co-occurrence characteristics of skeleton sequence informationcan be learned simply and effectively by utilizing a CNN model by utilizing the global aggregation capability of the CNN. In the method, point-level features of each joint are learned independently, and then the features of each joint are regarded as channels of a convolution layer to learn hierarchical symbiotic features. Most importantly, in a designed single pedestrian behavior recognition joint network structure, multi-part limb network features are fused into single pedestrian motion features to enhance behavior recognition of a single pedestrian. Besides, in the designed crowd interaction behavior recognition network, the characteristics of the group behaviors are enhanced by utilizing the characteristics of the single person behaviors, and the group behaviors relate to activities such as embracing and handshaking of behaviors of multiple persons.

Description

technical field [0001] The invention belongs to the field of deep learning for extracting facial features by deep neural network, relates to technologies such as neural network and pattern recognition, and particularly relates to a training and testing method for pedestrian and crowd behavior recognition models based on human skeleton information. Background technique [0002] The analysis of human actions such as action recognition and detection is one of the fundamental and challenging tasks in computer vision. In the behavior recognition technology with the human body as the main research object, most motion recognition methods use the method of target segmentation, but limited by factors such as the number of human bodies in the image and the size of the target, the effect is not ideal, resulting in unsatisfactory follow-up recognition work . Therefore, many research works often omit the process of moving target detection, and directly extract behavioral features from t...

AI Technical Summary

Problems solved by technology

Interference factors such as illumination changes, camera motion, and target occlusion in real scenes make this type of behavior recognition method less robust and less accurate.

Images