A pedestrian search method and device based on a priori candidate box selection strategy

Abstract

The invention discloses a pedestrian search method and device based on a prior candidate frame selection strategy. The method comprises the following steps: constructing a pedestrian candidate frame eigenvector according to the length and width of all pedestrian boundary frames in the training set, including two elements of the aspect ratio and the length; through a k-means + + algorithm, selecting the initialization clustering center; using the k-means algorithm for iterative clustering, and obtaining a priori candidate box after clustering; inputting the pedestrian images in the training setinto a pre-defined pedestrian search network, generating the candidate pedestrians by using a priori candidate boxes and recognizing the pedestrian identity, and obtaining the trained pedestrian search network by training; training pedestrian probes and surveillance scene images through the pedestrian search network to obtain pedestrian probes features and candidate pedestrian features in the surveillance image, and by calculating the similarity of the two features, locating the pedestrian probes in the surveillance image and its specific location. The invention can effectively improve the precision and speed of pedestrian search.

Description

technical field [0001] The invention belongs to the field of robot vision technology and intelligent monitoring, and specifically relates to a pedestrian search method and device based on a priori candidate frame selection strategy; by constructing the pedestrian candidate frame feature vector and designing a new distance measurement function, and using the bag of words model to The feature vector clustering of pedestrian candidate frames provides suitable initial candidate frames for pedestrian search, which can further reduce the regression time of pedestrian detection and improve the accuracy of pedestrian recognition in the later stage. Background technique [0002] Pedestrian search is a key technology that simultaneously solves pedestrian detection and pedestrian re-identification, and can be applied to human-computer interaction, intelligent monitoring, and video analysis. However, the performance of the pedestrian detection algorithm is still limited, and the resulti...

AI Technical Summary

Problems solved by technology

However, this method directly migrates the target detection network to pedestrian search, and the target size and pedestrian size in target detection are quite different. Since the two problems themselves have different scenarios, this type of method will introduce a large error

Another way (H.Liu, J.Feng, Z.Jie, K.Jayashree, B.Zhao, M.Qi, J.Jiang and S.Yan, “Neural person search machines,” in IEEE InternationalConference on Computer Vision (ICCV), 2017.) Use the information and attention mechanism of pedestrian probes to continuously match pedestrian probes and monitoring scene images. Compared with the former, the detection accuracy of this method is improved, but its algorithm efficiency is low. Pedestrians need to traverse the entire monitoring scene library, which is not suitable for actual monitoring scenarios

Images