Human body target tracking method applicable to depth image

Abstract

The invention provides a human body target tracking method applicable to a depth image. The human body target tracking method is characterized in that firstly a target template updating method is designed based on threshold judgment by using a human body target depth value; secondly, whether tracking is interfered by an obstacle to enable a tracking frame to jump or not is judged through calculating the distance between a predicted position of a Kalman filter and a mass center position of the current tracking frame, an obstacle shielding method is designed, and interference of an obstacle region for tracking is eliminated; then re-tracking is realized according to a method of human body target detection; and finally, a shielding obstacle removing mechanism is designed based on threshold judgment, a depth camera is enabled to be changed in viewing angle, and shielding of the obstacle region before the viewing angle is changed is removed. The human body target tracking method enables the success rate of human body target tracking in the depth image to be significantly improved.

Description

technical field [0001] The invention belongs to the field of computer vision, and relates to a method for tracking human body objects in depth images. Background technique [0002] In the field of modern artificial intelligence such as intelligent robots, intelligent monitoring and unmanned vehicles, human target detection and tracking is a key technical link, which plays a vital role in the control decision-making of intelligent equipment. Therefore, human target detection and tracking has always been a research hotspot in the field of computer vision. [0003] Depth information sensors represented by Microsoft Kinect can acquire depth image sequences with higher resolution and frame rate. Depth images have the advantages of providing distance information and strong anti-light interference ability. irreplaceable role. [0004] Kalman Filter is a recursive filtering algorithm based on the estimated information mathematical model. According to the state model and observati...

AI Technical Summary

Problems solved by technology

However, due to the particularity of the depth image, the combination method has the following problems when it is applied to the depth image: 1) When the human target is close to the obstacle, the tracking frame will jump to the obstacle area and cause tracking failure, and the traditional combined Kalman filter The MeanShift method cannot implement re-tracking

2) When the human target is farther away from the depth camera than at the initial tracking moment, and after the depth value changes, when an obstacle closer to the depth camera appears in the pixel area near the human target, the tracking frame will jump to the obstacle area, causing the tracking to fail, and The traditional MeanShift method combined with Kalman filter cannot achieve re-tracking

Images