Kernel correlation filtering target tracking method suitable for pedestrian following of mobile robot

Abstract

The invention discloses a kernel correlation filtering target tracking method suitable for pedestrian following of a mobile robot. The kernel correlation filtering target tracking method comprises thesteps of utilizing SVM pedestrian classifier based on OpenCV to detect and initialize target position and target area; constructing a training sample according to the target area of the current frame, and performing multi-feature extraction and weighted fusion to obtain a feature vector; constructing a ridge regression model classifier for target tracking by taking a kernel autocorrelation cyclicmatrix in a Fourier space as input and taking a regression value as output, and calculating to obtain a learning weight coefficient; reading in the next frame, constructing a detection sample according to the target position of the previous frame, and forming a cross-correlation matrix with the training sample; establishing a scale pyramid and combining bilinear interpolation to obtain target detection areas of different scale models, calculating to obtain the maximum response and updating the target position; and training and updating the target tracking ridge regression model classifier again. According to the method, the target can be effectively captured, multi-level scale adaptive transformation is achieved, and good robustness and real-time performance are achieved.

Description

technical field [0001] The invention relates to a kernel correlation filtering target tracking method suitable for pedestrian following of a mobile robot, and belongs to the technical field of visual target tracking. Background technique [0002] In the era of artificial intelligence, robotics is an important symbol of application development. Among them, the mobile robot is a representative intelligent robot technology, which provides a lot of convenience for people's daily work and life, integrates into all walks of life such as large shopping malls, hotels, catering and transportation, and improves people's quality of life. [0003] Object tracking technology for mobile robots is one of the research hotspots in image processing and machine vision. In the application scenario of pedestrian following, it will be affected by problems such as poor real-time performance, changes in light brightness, attitude changes, scale changes, target occlusion, and complex backgrounds. ...

AI Technical Summary

Problems solved by technology

[0006] The technical problem to be solved by the present invention is to overcome the lack of anti-occlusion performance and the inability to adapt to the tracking scene of scale transformation in the traditional high-speed kernel correlation filtering KCF algorithm, and to provide a kernel correlation filtering target suitable for pedestrians following by mobile robots Tracking method, in order to solve the above problems and ensure the real-time processing of pedestrian following, a weighted fusion of FHOG features, CN features and LBP features is proposed, and a method of increasing the scale factor to adapt to the scale transformation is proposed.

Images