Robust tracking method of target in airport monitoring video

Abstract

The invention relates to a robust tracking method of target in an airport monitoring video, which combines singular value decomposition (SVD) and particle filtering target tracking technology. The method comprises the following steps: step 100: initiating, generating particle swarm containing N particles, carrying out SVD for the tracked target, and establishing template model; step 200: forecasting state by forecasting position of the target in current frame based on position of the target in a prior frame; step 300: updating particle weight, carrying out SVD for candidate image area, establishing feature vector, and comparing similarity of the candidate area and the template by using least squares criterion to be weight of the particle; step 400: re-sampling particle, keeping particles with larger weight, discarding particles with smaller weight, enabling particles with larger weight to derive new particles to keep total number N of particles unchanged, and transferring new particles and the next frame of image to step 200 for circulation operation; and step 500: outputting result, and weighting all particles and weight of all particles to obtain position of the target in current frame. The invention solves problems of complicated scene, frequent illumination change, serious interference caused by background analogue and the like of existing airport monitoring video, and meets requirements of robustness and accuracy in monitoring and tracking process.

Description

technical field

[0001] The invention relates to a target tracking method, in particular to a target tracking method in an airport monitoring video image. Background technique

[0002] At present, the main means of surveillance of surface moving targets at airports is to rely on the surface surveillance radar, and at the same time use the multilateration (Multilateration, MLAT) surface surveillance system as an auxiliary. The current field surveillance radar has obvious shortcomings, such as expensive, insufficient resolution (cannot detect foreign objects on the runway), and radar coverage in the blind area. If the data fusion of multiple field surveillance radars is used, the complete surveillance of the entire airport can be realized, but the system construction cost will be greatly increased, and there are not many mature applications for field surveillance radar data fusion. Compared with field surveillance radar, multipoint positioning has the advantages of low cost, a...

Examples

Embodiment Construction

[0029] figure 1 It is an overall flow chart of the present invention, specifically:

[0030] Step 100, initialization, adding Gaussian noise with the input value of the target state to generate Particle swarm of particles, and perform singular value decomposition on the target to be tracked, and select the top K maximum values ​​to establish a template model;

[0031] Step 200, state prediction, predicting the position of the target in the current frame from the position of the target in the previous frame according to the uniform motion model;

[0032] Step 300, update the particle weight, perform singular value decomposition on the candidate image region, construct a feature vector, and use the least squares criterion to measure the similarity between the candidate region and the template as the weight of the particle;

[0033] Step 400, Particle resampling, keep the original weighted particles, discard the small weighted particles, and derive new particles from the heavy...

technical field

[0001] The invention relates to a target tracking method, in particular to a target tracking method in an airport monitoring video image. Background technique

[0002] At present, the main means of surveillance of surface moving targets at airports is to rely on the surface surveillance radar, and at the same time use the multilateration (Multilateration, MLAT) surface surveillance system as an auxiliary. The current field surveillance radar has obvious shortcomings, such as expensive, insufficient resolution (cannot detect foreign objects on the runway), and radar coverage in the blind area. If the data fusion of multiple field surveillance radars is used, the complete surveillance of the entire airport can be realized, but the system construction cost will be greatly increased, and there are not many mature applications for field surveillance radar data fusion. Compared with field surveillance radar, multipoint positioning has the advantages of low cost, a...

Examples

Embodiment Construction

[0029] figure 1 It is an overall flow chart of the present invention, specifically:

[0030] Step 100, initialization, adding Gaussian noise with the input value of the target state to generate Particle swarm of particles, and perform singular value decomposition on the target to be tracked, and select the top K maximum values ​​to establish a template model;

[0031] Step 200, state prediction, predicting the position of the target in the current frame from the position of the target in the previous frame according to the uniform motion model;

[0032] Step 300, update the particle weight, perform singular value decomposition on the candidate image region, construct a feature vector, and use the least squares criterion to measure the similarity between the candidate region and the template as the weight of the particle;

[0033] Step 400, Particle resampling, keep the original weighted particles, discard the small weighted particles, and derive new particles from the heavy...

Images