Target tracking method and device

Abstract

The embodiment of the invention discloses a target tracking method and a target tracking device, which are used for reducing the calculation complexity of target tracking and have stronger generalization ability. The method comprises the following steps: acquiring a first frame image to be processed from video data acquired by a camera; performing target detection on the first frame image to generate a target detection result, the target detection result including a first target detected from the first frame image; performing motion estimation on the first target by adopting a Kalman motion model to generate a target tracking result, the target tracking result comprising a prediction position of the first target in a next frame image relative to the first frame image; and estimating the motion state of the first target according to the target detection result and the target tracking result.

Description

technical field [0001] The present invention relates to the field of computer technology, in particular to a target tracking method and device. Background technique [0002] In the target visual tracking technology, the movement state of the target can be tracked in real time through a single camera. The target tracking algorithm used in the prior art is mostly divided into two processes. The first step is to independently detect the visual target in each frame of the image. For example, the visual target can be pedestrians, vehicles, etc. The commonly used target detection algorithm has deformable parts Model, Regions with Convolutional Neural Network, RCNN, Faster Regions with Convolutional Neural Network, Faster RCNN, etc. The second step is to correlate the visual targets detected in consecutive frames of images. According to the similarity calculation between target features, a similarity matrix or loss matrix can be obtained, and the target matching result is generate...

AI Technical Summary

Problems solved by technology

[0003] In the prior art, target detection algorithms all use complex motion models, which lead to high computational complexity, making the processing speed of the target tracking algorithm around 1 hertz (Hz) or even lower, that is, the target tracking algorithm processes frame The rate can only be processed once per second, which is far from meeting the needs of real-time usage scenarios

As another example, algorithms based on deformable component models, RNN, and Faster RCNN in the prior art have weak generalization ability and are easy to overfit to certain types of motion modes, making it difficult to promote them to more stable and general motion modes.

Therefore, the main disadvantages of the target tracking algorithm in the prior art are that the computational complexity is too high and the generalization ability is weak

Images