High-level semantic video behavior identifying method based on confusion matrix

Abstract

The invention discloses a high-level semantic video behavior identifying method based on a confusion matrix, and mainly solves the problems of low identification ratio caused by confusion in the priorart. Realization steps of the high-level semantic video behavior identifying method are as follows: 1) extracting dense tracks of behavior videos, and acquiring bottom-level features of the tracks; 2) identifying behaviors by utilizing the bottom-level features to acquire the confusion matrix; 3) aiming at the confusion matrix, defining a high-level semantic list which can distinguish confusion behaviors; 4) associating the high-level semantic list with the behavior video data, training a corresponding discriminative classifier for each high-level semantic meaning, and cascading score discrimination values of behavior videos under the all discriminative classifiers to acquire high-level semantic feature vectors; and 5) acquiring membership grades of the bottom-level features under linearSVM (Support Vector Machine) classifiers, combining the membership grades with the high-level semantic feature vectors, and training LSVM (Linear Support Vector Machine) classifiers to carry out behavior identification. According to the high-level semantic video behavior identifying method based on the confusion matrix disclosed by the invention, confusion behaviors can be pointedly corrected, sothat the correctness of the behavior identification is improved, and the high-level semantic video behavior identifying method can be applied to video monitoring.

Description

technical field [0001] The invention belongs to the technical field of video image processing, and in particular relates to a video behavior recognition method, which can be used for video monitoring. Background technique [0002] In recent years, the increasingly important academic value, economic benefit and social value of behavior recognition and video classification have attracted the attention of many scholars and become a challenging research topic in the field of computer vision. New human-computer interaction, virtual reality, video coding and transmission, game control and many other aspects have broad application prospects. With the deepening of research work in the field of video behavior recognition, researchers have found that low-level features have very limited ability to describe complex video behaviors, and research hotspots have gradually shifted from the design and extraction of new low-level features to the definition of high-level semantics and the cons...

AI Technical Summary

Problems solved by technology

However, the learning of objects and poses requires additional learning of object detectors and pose detectors, and object detectors need to be implemented on the basis of accurate object detection, which is difficult to apply to complex human behaviors.

Images