Target detection and tracking method based on radar and vision fusion in dynamic background environment

Abstract

The invention discloses a method for radar and vision fused target detecting and tracking in a dynamic background environment. The method includes steps: 1) fixedly arranging radar and an image acquisition device in advance, and adopting the image acquisition device for acquiring background images according to different focal lengths and different angles and establishing a background image library; 2) when a target is detected by the radar, controlling and regulating the image acquisition device to acquire current images; 3) searching background images matched with the current images, and extracting foreground images corresponding to the current images according to the searched background images; 4) subjecting the foreground images to target detecting and tracking. The method is suitable for conditions of complex background environments, high target moving speed, image zooming and the like to realize target detecting and tracking and has advantages of simplicity in implementation, low cost, high environment adaptability and effectiveness in detecting and tracking.

Description

Technical field [0001] The invention relates to the technical field of dynamic target detection and tracking, in particular to a target detection and tracking method of radar and vision fusion in a dynamic background environment. Background technique [0002] Dynamic target detection and tracking are usually divided into two types of conditions: static background environment and dynamic background environment. For dynamic background environment conditions, most of them use video functions to analyze image content. The analysis methods mainly include optical flow method and mean value. Drift, particle filter, Kalman filter, TLD, etc. For example, Chinese patent application 201310059457.5 discloses a vision-based dynamic target tracking and positioning method for drones. The video processing method is to detect and track the dynamic target in the image. The basic method used in the detection part is the frame difference method, and the tracking part The method used is continuous t...

AI Technical Summary

Problems solved by technology

[0003] (1) Using the video function alone to analyze the image content makes the design of the realization method complex and difficult, and at the same time, the environment adaptability is poor, and it is easily disturbed by the external lighting environment or the complex background environment, such as when the moving target and its surroundings When the background is similar, the tracking accuracy will drop, or even fail completely, and the target cannot be correctly distinguished;

[0004] (2) The tracking effect of slower moving objects is better. When it is applied to the detection and tracking of faster moving objects, the faster moving objects will cause the feature points of two adjacent frames to have a smaller matching degree. The global motion vector of the current image cannot be calculated, that is, the foreground area cannot be extracted by subtraction of adjacent frames, and the phenomenon of tracking loss is prone to occur;

[0005] (3) At present, target detection and tracking are usually completed under the condition of a fixed focal length field of view. Once applied to the situation where the focal length of the image sensor is variable, the zooming process will also result in a small matching degree of feature points between adjacent frames. , it is difficult to calculate the global motion vector of the current image, and it is easy to lose the key features of the target. In addition, to obtain better results, a hardware system with higher processing performance is required, and the image will undergo severe deformation during the zooming process. Therefore, the current It is usually impossible to detect and track the target during the zoom process;

[0006] (4) The image information needs to be analyzed with the help of thermal imaging equipment, etc. Thermal imaging equipment is expensive, which makes the cost of detection and tracking relatively high

Images