3D moving target detection method based on 6-freedom degree scene flow clustering

Abstract

The invention provides a 3D moving target detection method based on 6-freedom degree scene flow clustering. The method comprises steps of acquiring an aligned color image and a depth image of a scene by use of a depth camera; constructing a 6-freedom degree scene flow estimation energy function; carrying out optimal solving of the energy function; according to the definition of the scene flow, using a rotation vector and a translation vector to calculate the scene flow; carrying out initial analysis according to the information of the scene flow, and determining general quantity of moving targets; extracting moving feature information according to the scene flow and acquiring a feature vector of each point; and using the ISODATA algorithm to carry out cluster analysis and extracting a moving target. According to the invention, by use of neighborhood constrains and combining brightness constant and depth constant constrains, the data item is constructed; smooth constraining is performed on the rotation vector and the translation vector by use of all variation smooth; and after solving of the scene flow, by using the ISODATA algorithm, the cluster analysis is performed on the scene flow, and a 3D moving target is extracted.

Description

technical field [0001] The invention relates to a 3D moving target detection method. Background technique [0002] An important goal in machine vision is to realize the detection of moving objects, and even the analysis and understanding of images or videos. Moving object detection is an important direction in the field of machine vision, and has a wide range of applications, such as video surveillance, human-computer interaction, scene analysis and other basic technologies involving image or video analysis and understanding. The classic methods of static background moving target detection include background subtraction method, frame difference method and optical flow method. The background static subtraction method extracts moving objects through background modeling, but this method is particularly sensitive to environmental changes. The inter-frame difference method is to make a difference between two or three adjacent frames to the moving target, but this method require...

AI Technical Summary

Problems solved by technology

The background static subtraction method extracts moving targets through background modeling, but this method is particularly sensitive to environmental changes

The inter-frame difference method is to make a difference between two adjacent frames or three frames to reach the moving target, but this method requires obvious pixel changes when the object is moving, otherwise it will cause missed detection and other phenomena

The optical flow method is an important tool for target detection. This method calculates the optical flow vector and detects the moving target according to the difference between the moving object and the background motion vector. In the dynamic background, this method can also achieve better results, but when When the object motion is moving towards the camera direction, it will cause target detection error

None of the above methods has a two-dimensional detection method for the depth information of the scene. 3D moving object detection can detect moving objects more accurately, and has important applications in areas such as vehicle assisted driving.

Images