A human-vehicle separation method based on RFCN

Abstract

The invention provides a human-vehicle separation method based on RFCN, and the method comprises the steps: enabling a computer to directly obtain a video frame image of a monitoring image from a camera, and carrying out the simple preprocessing of the image; detecting the whole image by using a pre-trained deep learning detection model, identifying and positioning all people and vehicle targets in the image, and outputting an external rectangle, category and confidence coefficient of the target; the computer performs post-processing on the detection result, removes an external rectangular frame with low confidence coefficient and high overlapping degree, and outputs the external rectangular frame as a final result; and comprehensively analyzing the continuous video frame images, and judging the motion state of the target and the time of entering the scene to serve as reference conditions for judging and triggering warning. According to the RFCN-based human-vehicle separation method provided by the invention, the judgment of the target category and the positioning regression of the position can be accurately realized, a moving target can be detected, a static target can also be detected, and non-human and vehicle targets can be accurately filtered out.

Description

technical field [0001] The invention belongs to the field of video monitoring, and in particular relates to a method for separating people and vehicles based on RFCN. Background technique [0002] Along with the development of high technology and social progress, video surveillance systems are widely used in more and more occasions. The camera alert function of the current monitoring system cannot meet various complex scenes and more and more different needs. It is mainly reflected in that the warning scene requires the monitoring system not only to accurately detect the targets in the scene, but also to automatically identify the category and location of the target. The main monitoring targets in daily alert scenarios are people and vehicles, so the target categories to be detected are also people or vehicles, so a method is needed to accurately identify and classify the detected targets. The existing security cameras use cascade detection, support vector machine or simpl...

AI Technical Summary

Problems solved by technology

The existing security cameras use cascade detection, support vector machine or simple BP neural network and other detection and classification methods to basically realize the category judgment function of the detection target, but these methods are suitable for limited scenarios, and the accuracy of target category detection It is relatively low, especially for the recognition and classification of small targets. In addition, the preprocessing method using background modeling technology can only detect moving targets, which cannot meet the needs of detecting stationary targets in the scene, and it is difficult to widely popularize and apply.

Images