Automatic detection and tracking method based on target adaptive projection

Abstract

The invention discloses an automatic detection and tracking method based on target adaptive projection. The method comprises the following steps: reading to-be-tracked video sequence information, andperforming RGB-to-Gray color space conversion on each frame of image; solving a foreground target mask in the first M frames by utilizing a weighted sliding variance method; adaptively selecting the maximum mask of the foreground target area; carrying out horizontal and vertical projection on the mask; searching the projection result in the vertical and horizontal directions, and determining the center of the optimal target through accumulated pixels and the proportion; performing round-trip detection in the vertical and horizontal directions from the center position, determining the optimal target scale through accumulated pixels and the proportion, and thus determining an ROI (target area); and finally, tracking the target by using a centroid method to finish an automatic target detection and tracking process. The method is easy to implement and clear in process, and the influence of background noise can be effectively avoided in the ROI extraction process through the self-adaptive projection method.

Description

technical field [0001] The invention belongs to the field of industrial computer vision and relates to the problems of background modeling, target detection, target tracking and initial ROI calculation. Background technique [0002] In the field of computer vision, object detection and object tracking are two hot research issues. The current research process on these two issues is slowly transitioning from traditional methods to deep methods. In most research works, object detection and object tracking are studied as two separate problems, and an ideal situation is that for a given video sequence, the algorithm can automatically frame the object and automatically track it. . Although deep learning methods have achieved good results in recent years, deep methods require large datasets to train models, and in real scenarios, data acquisition is often difficult. Therefore, the traditional method still plays an indispensable role, and in the traditional method, it is a very fe...

AI Technical Summary

Problems solved by technology

Although deep learning methods have achieved good results in recent years, deep methods require large data sets to train models, and in real scenarios, data acquisition is often difficult

Therefore, the traditional method still plays an indispensable role, and in the traditional method, it is a very feasible solution to use the background modeling method for target detection, but the output of the background modeling may contain noise, which will to some extent Affect the accuracy of the extracted target ROI

Images