Pedestrian detection and tracking method based on head-shoulder contour and BP neural network

Abstract

The invention proposes a pedestrian detection and tracking method based on a head-shoulder contour and a BP neural network. The method comprises the steps: firstly extracting a moving human body target in a video sequence through employing an adaptive mixed Gaussian background updating algorithm, and improving the background estimation precision through changing a learning factor of a mixed Gaussian model; secondly extracting an initial contour of an original target through employing a Canny operator, and carrying out contour clustering through combining a Mean shift algorithm, so as to obtain a completer body contour; thirdly building a head-shoulder contour model through combining a head-shoulder width-height ratio of a human body, extracting a head-shoulder contour characteristic vector, inputting the characteristic vector of the head-shoulder contour model into the BP neural network, clustering a plurality of human body head-shoulder models, and carrying out human body recognition; and finally tracking a detected pedestrian target through employing a particle filter. The method avoids misjudgment and wrong judgment because of the incompletion of a recognition target, improves the recognition accuracy of the pedestrian target, and reduces the calculation amount.

Description

technical field [0001] The invention belongs to the technical field of moving target detection and tracking, in particular to a pedestrian detection and tracking method based on a head-shoulder profile and a BP neural network. Background technique [0002] The detection, recognition and tracking of human targets is one of the hot research issues in the field of computer vision recognition, and its accuracy affects the smooth progress of follow-up work such as target tracking, behavior recognition and analysis. [0003] Human body recognition judges whether a moving target is a human target through the acquired information features. People such as N.Dalal use the whole human body as a recognition model, by calculating the HOG (HistogramsofOrientedGradients, direction gradient histogram) feature of the model, and combining SVM (Support-VectorMachines, support vector machine) classifier to realize human body recognition; Kuno people etc. use projection histograms to analyze ta...

AI Technical Summary

Problems solved by technology

However, since the human body is a non-rigid object, the posture in motion is uncertain, and the complete human body is easily blocked by external objects, which will affect the accuracy of the above method of human body recognition

Images