Optimization of particle filter tracking algorithm based on weakly supervised attribute learning

This study proposes an optimization method for particle filter tracking algorithm to solve the issues of low recognition efficiency and poor tracking accuracy faced by existing target tracking algorithms in complex environments. This method combines weakly supervised learning with energy function optimization to raise the efficiency of image feature annotation in object detection models. Besides, to raise the robustness and accuracy of target tracking algorithms in complex environments, an improved particle filter tracking method based on accelerated robust feature matching is proposed. The simulation results show that compared with recurrent neural networks, this method reduces the recognition errors of target center point and target size by 36.61% and 37.53% respectively during the daytime. Compared with the support vector machine model, this method reduces recognition errors by 23.01% and 28.43%, respectively. In the case where the target is obstructed, the tracking accuracy of the raised method is as high as 0.95. The outcomes denote that the raised method has excellent robustness and target tracking accuracy, and can provide effective solutions for target tracking problems in complex environments.