Autonomous learning multi-target detection method based on hybrid classifier

Abstract

The invention provides an autonomous learning multi-target detection method based on a hybrid classifier. According to the method, only through frame selection of an interested detection target from a first frame of a video, a target detection classifier can be automatically initialized by a system according to the target acquired through frame selection, through continuously autonomous learning, detection performance of the classifier is gradually improved. The method is advantaged in that the random fern classifier and iterable SVM are combined and commonly act on detection classification of the target, tests on vehicles and pedestrians are carried out, and the relatively good effect is acquired.

Description

technical field [0001] The invention relates to a self-learning multi-target detection method based on a hybrid classifier, which belongs to the field of pattern recognition. Background technique [0002] Video surveillance is widely used in various fields such as community housing, transportation facilities, financial institutions, and public entertainment venues. As the main targets of surveillance, the detection of vehicles and pedestrians is an important issue in intelligent video processing. Disciplinary technology and broad market prospects have become the research and development focus of various research institutions and companies around the world. However, it is still a challenging task to achieve stable and accurate vehicle and pedestrian detection in a variety of dynamically changing scenes and under different levels of congestion. [0003] By analyzing the number and movement information of vehicles, we can have a clear understanding of traffic congestion and av...

AI Technical Summary

Problems solved by technology

However, since the Edgelet feature needs to be manually calibrated, the extraction of the feature is more complicated. For some complex curves, it is difficult to obtain the Edgelet feature that conforms to the human body curve through manual calibration.

The SIFT feature can achieve very good results for extracting the invariant features of the image, but there are still many defects, such as: sometimes the extracted feature points are relatively small, the real-time performance is average, and the target object with blurred edges cannot be extracted, etc.

Images