A pedestrian re-recognition and tracking method based on spatio-temporal context

Abstract

The invention relates to a pedestrian re-identification and tracking method based on spatio-temporal context, which comprises the following steps: training a Mask RCNN network; Using the trained MaskRCNN network to process the original picture set, the training set, the test set and the search set. Training convolution neural network with training set; The test set and the search set are processed using the trained convolution neural network to obtain a first preset number of pictures from the test set for reidentifying the target pedestrian. The invention uses the object detection algorithmand the instance segmentation algorithm to preprocess the picture, removes the background interference information, further improves the model precision, and improves the accuracy of the pedestrian re-identification method. At that same time, the invention solve the problem that the current pedestrian re-identification algorithm lacks the tracking function, and proposes the area prediction algorithm based on the walking speed, and combine the Mask RCNN to reduce the tracking calculation complexity, achieve real-time tracking, and improve the tracking efficiency.

Description

technical field [0001] The technical field of image processing of the present invention particularly relates to a pedestrian re-identification and tracking method based on spatio-temporal context. Background technique [0002] Person re-identification (Person re-identification), also known as pedestrian re-identification, is a technology that uses computer vision technology to determine whether a specific pedestrian exists in an image or video sequence. The pedestrian re-identification algorithm plays a very important role in the field of public security. It is applied to quickly retrieve and track targets in surveillance videos, so as to make up for the lack of identification in the absence of face information. The task of pedestrian re-identification is cross-camera retrieval, that is, objects appearing in one camera need to be retrieved from other cameras. [0003] At present, pedestrian re-identification algorithms can be mainly divided into two categories: one is image...

AI Technical Summary

Problems solved by technology

[0004] However, the DukeMTMC-reID database commonly used in the research of image-based pedestrian re-identification algorithms has only about 30,000 pictures (including more than 1,400 pedestrians), resulting in insufficient data for training image-based pedestrian re-identification algorithms. Convolutional neural network, which affects its re-identification accuracy; for video-based pedestrian re-identification algorithms, it needs to be trained in convolutional neural network and recurrent neural network, resulting in high complexity of such algorithms, and the required longer training time

[0005] Moreover, in some existing tracking methods, most of them use long short-term memory network (a kind of recurrent neural network) to predict the position of pedestrians, and this tracking method uses traditional manual feature extraction to extract pedestrian information, which is accurate. The rate is low, and the training takes a long time, resulting in low tracking efficiency

Images