Multi-target tracking method and system based on depth condition random field model

Abstract

The invention discloses a multi-target tracking method based on a depth condition random field model. The method comprises the steps of obtaining a multi-target tracking data set, carrying out data association on detection responses of any two continuous frames in all frames of the input video sequence; track sheet, generating a vertex according to the time relationship between any two track pieces in the obtained track piece set; calculating the matching degree between two track pieces corresponding to each vertex in the vertex set; determining a difficult vertex pair set according to a timerelationship and a position relationship between any two vertexes in the vertex set; and obtaining the appearance characteristics and position information of each difficult vertex pair in the obtaineddifficult vertex pair set, combining the appearance characteristics and motion characteristics of the difficult vertex pair into a difficult vertex pair characteristic vector, and inputting each difficult vertex pair characteristic vector into an LSTM network. According to the method, the correlation of real data in the multi-target tracking process can be effectively embodied, and the accuracy of a tracking result is high.

Description

technical field [0001] The invention belongs to the technical field of pattern recognition, and more specifically, relates to a multi-target tracking method and system based on a depth conditional random field model. Background technique [0002] Multiple object tracking (MOT) has attracted much attention in the field of computer vision because of its academic potential and commercial value. The main task of MOT is to retrieve the motion trajectory of the target of interest in a given video sequence, calibrate the position of the target of interest, and identify the identity (ie ID) of each target of interest. The target of interest can be a pedestrian. , vehicles, animals, and even different components of an object. [0003] Most of the existing multi-target tracking methods are based on the data association model of detection, that is, after the target detection response of the step-by-step frames in the video is given, the target tracking is transformed into a data assoc...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the above defects or improvement needs of the prior art, the present invention provides a multi-target tracking method and system based on the depth-conditional random field model, the purpose of which is to solve the existing classic multi-target tracking method based on CRF modeling. Due to the complexity of the multi-target tracking problem, the specific distribution cannot effectively reflect the correlation of real data, which will reduce the accuracy of the target tracking results, and the difficulty of solving the parameters of the CRF model will not only cost a lot of Manpower and computing power, and the technical problems that the optimal performance of the CRF model cannot be exerted, and its reasoning algorithm is difficult to combine with the neural network, so it is impossible to form an end-to-end trainable deep neural network, so that deep learning cannot be used Strong ability to learn technical issues

Images