Mobile Robot Following Research Based On Regional Tracking And Human Target Re-matching

With the aggravation of the aging population and the deepening of the demand for convenience in human daily life,it is increasingly urgent to introduce mobile robots in airports,shopping malls and other occasions to reduce material and labor costs.The following of human targets by mobile robots under the scene of interference is mainly studied in this paper.During the following process,it is easy to be interfered by illumination change,scale change,deformed targets and occlusion,resulting in the loss or mistracking of the target.In order to improve the anti-interference performance of mobile robot tracking,a new model updating strategy,depth feature and scale estimation based on improved kernel correlation filter are proposed in this paper.In order to adjust the model updating strategy in kernel correlation filters algorithm,the occlusion process is analyzed in stages and the rules are summarized.Kernel correlation filter method has a high computing speed,but its tracking accuracy is limited by the feature of manual selection.In this paper,due to the limited size of the existing target tracking data set,the features of each layer of the classical classification model vgg16-net are analyzed,three depth features that are beneficial to target tracking are selected and the model is fine-tuned.Discrete wavelet transform is used to fuse the three depth features and the fused feature is used to replace the histogram feature of gradient direction in the kernel correlation filter method to complete the position filter training.By means of scale pyramid,bilinear interpolation and kernel correlation filter framework,the scale filter is trained to reduce the influence of scale change.In this paper,qualitative and quantitative analysis of the improved algorithm in this paper is completed with the help of OTB2013 benchmark data set.Experimental results show that the improved algorithm in this paper has been improved in precision and success rate compared with the kernel correlation filter algorithm.As time goes by,errors accumulate during target tracking.In this paper,a method of human target re-matching based on lidar and vision is designed.Lidar is used to identify legs and the modified saliency model is designed to obtain the fine detection area of human target.The feature of histogram of gradient direction based on the coarse and fine detection area is extracted and the candidate pedestrian area is obtained by using the classifier based on INRIA data set training in this paper.The candidate pedestrian area and the target pedestrian area initialized in the first frame image are sent to the adjusted siamese network trained based on the Market-1501 data set to complete the human target re-matching.In order to better obtain the position of the tracked target,a trigger mechanism based on error accumulation is designed in this paper,which combines the human target re-matching strategy with the region-based target tracking method.The experimental results in the real scene of mobile robot fixed position show that the algorithm is more robust to interference.The collection of target positions in each frame of image forms the motion trajectory.During the trajectory tracking process of the mobile robot,there is deviation between the actual pose and the expected pose.We use the idea of inversion controller design to select lyapunov function and design the control law.Aiming at the uncertain parameters in the control law,the fuzzy control method is adopted in this paper to make the control law more in line with the research requirements.In this paper,the trajectory tracking simulation effect under Matlab is good and the trajectory tracking experiment and the following experiment of the mobile robot under the interference scene are completed in the real scene.