Research And Implementation Of Pedestrian-oriented Multiple Object Tracking Technology

At present,multi-target tracking technology has been widely used in all aspects of our lives.The scene involved in the multi-target tracking task is complex:there are some phenomena,such as the entry and disappearance of the target,the frequent occlusion of the target and so on.The multi-target tracking algorithm will cause the phenomenon of tracking error when the target is occluded;because the target is missed,the target is false disappearance,and when tracking the target again,it will cause the phenomenon of tracking error.Therefore,how to achieve the correct correlation between multi-objectives in complex scenes is still a difficult problem.Therefore,starting from the problem of how to correctly realize MOT in complex environment,this thesis proposes a new pedestrian-oriented multi-target tracking algorithm.The main research contents are as follows:1.In order to solve the problem that the performance of the algorithm degrades when the pedestrian is occlusion,an occlusion-oriented MOT algorithm is proposed(OOMOT).First of all,in view of the fact that the existing motion estimation methods can't effectively learn more complex information between motion information,a motion estimation algorithm based on convolution network is proposed.The convolution layer uses dilated convolution,which can ensure that the neuron only learns the data information before the current moment,does not depend on the future information,and takes into account the timing relationship of data input.Then in view of the existing measurement methods,the weight relationship between appearance model and motion model measurement results is ignored.Unable to effectively weigh the measurement results of appearance features and motion features,an adaptive feature measurement algorithm is proposed.Finally,experiments are carried out on the data set MOT 16,and the experimental results can reflect the occlusion-oriented MOT algorithm proposed in this thesis,which can achieve the correct association of pedestrians in the case of occlusion.2.In order to solve the problem that long-time consistent tracking can't be achieved after target missed detection,a long-time multi-target tracking algorithm based on trajectory verification is proposed(LTMOT).The long-time multi-target tracking algorithm takes the trajectory verification algorithm as the core.The main purpose of this algorithm is to check the new trajectory to prevent the track number of the current target from changing when the target is identified again after missing detection.In the results of the dataset MOT 16,it can be found that the combination of the trajectory verification algorithm and the occlusion-oriented MOT algorithm can effectively achieve long-time tracking when the target is identified again after missed detection.3.A multi-pedestrian tracking and monitoring system is realized.The system first carries out data preprocessing according to the received video data,then carries out target detection of multi-frame images according to the preprocessed data,and then realizes the association between targets based on the results of target detection.finally,the results of multi-target tracking can be fed back to the user and displayed in the front-end interface.