In order to solve the problem of low accuracy of video-based person re-identification caused by factors such as occlusion, background interference, and person appearance and posture similarity in video surveillance, a video-based person re-identification method of Evenly Sampling-random Erasing (ESE) and global temporal feature pooling was proposed. Firstly, aiming at the situation where the object person is disturbed or partially occluded, a data enhancement method of evenly sampling-random erasing was adopted to effectively alleviate the occlusion problem, improving the generalization ability of the model, so as to more accurately match the person. Secondly, to further improve the accuracy of video-based person re-identification, and learn more discriminative feature representations, a 3D Convolutional Neural Network (3DCNN) was used to extract temporal and spatial features. And a Global Temporal Feature Pooling (GTFP) layer was added to the network before the output of person feature representations, so as to ensure the obtaining of spatial information of the context, and refine the intra-frame temporal information. Lots of experiments conducted on three public video datasets, MARS, DukeMTMC-VideoReID and PRID-201l, prove that the method of jointing evenly sampling-random erasing and global temporal feature pooling is competitive compared with some state-of-the-art video-based person re-identification methods.

Aiming at the problem of poor pedestrian detection effect in complex scenes, a pedestrian detection algorithm based on improved Mask R-CNN framework was proposed with the use of the leading research results in deep learning-based object detection. Firstly, K-means algorithm was used to cluster the object frames of the pedestrian datasets to obtain the appropriate aspect ratio. By adding the set of aspect ratio (2:5), 12 anchors were able to be adapted to the size of the pedestrian in the image. Secondly, combined with the technology of fine-grained image recognition, the high accuracy of pedestrian positioning was realized. Thirdly, the foreground object was segmented by the Full Convolutional Network (FCN), and pixel prediction was performed to obtain the local mask (upper body, lower body) of the pedestrian, so as to achieve the fine-grained detection of pedestrians. Finally, the overall mask of the pedestrian was obtained by learning the local features of the pedestrian. In order to verify the effectiveness of the improved algorithm, the proposed algorithm was compared with the current representative object detection methods (such as Faster Region-based Convolutional Neural Network (Faster R-CNN), YOLOv2 and R-FCN (Region-based Fully Convolutional Network)) on the same dataset. The experimental results show that the improved algorithm increases the speed and accuracy of pedestrian detection and reduces the false positive rate.

Focusing on the person re-identification in non-overlapping camera views and the high dimensional feature extracted from the images, a person re-identification method based on block sparse representation was proposed. The Canonical Correlation Analysis (CCA) was taken to carry out the feature projection transformation, and the curse of dimensionality caused by high dimensional feature operation was avoided by improving the feature matching ability, and the feature vectors in a probe image were made to be probably linear with the corresponding gallery feature vectors in the learned projected space of CCA transformation. A person re-identification model was also built with block structure feature of pedestrian dataset, and the associated optimization problem was solved by utilizing the alternating direction framework. Finally, the residues were used to deal with the person in the probe set to be identified and the index of the minimum value in the residues was regarded as the identity of the person. Several experiments were conducted on public datasets such as PRID 2011, iLIDS-VID and VIPeR. The experimental results show that the Rank1 value of the proposed method on three experimental datasets reaches 40.4%, 38.11% and 23.68%, respectively, which is significantly higher than that of Large Margin Nearest Neighbor (LMNN) method, and the matching rate of it on Rank-1 is also much bigger than that of LMNN method; besides, the overall performance of it is better than the classical algorithms based on feature representation and metric learning. The experimental results verify the effectiveness of the proposed method on person re-identification.

Person Re-Identification (Re-ID) under non-overlapping multi-camera is easily affected by illumination, posture, and occlusion, and there are image mismatches in the experimental process. A Re-ID method based on siamese network and reranking was proposed. Firstly, a pair of pedestrian training images were given, a discriminative Convolutional Neural Network (CNN) feature and similarity measure could be simultaneously learned by the siamese network to predict the pedestrian identity of the two input images and determine whether they belonged to the same pedestrian. Then, the k-reciprocal neighbor method was used to reduce the image mismatches. Finally, Euclidean distance and Jaccard distance were weighted to rerank the sorted list. Several experiments were performed on the datasets Market1501 and CUHK03. The experimental results show that the Rank1 (the probability of matching successfully for the first time) reaches 83.44% and mAP (mean Average Precision) is 68.75% under Single Query on Market1501. In the case of single-shot on CUHK03, the Rank1 reaches 85.56% and mAP is 88.32%, which are significantly higher than those of the traditional methods based on feature representation and metric learning.

According to the difference between normal users visiting patterns and abnormal ones, a new method to detect applicationlayer Distributed Denial of Service (DDoS) attack was proposed based on IP Service Request Entropy (SRE) time series. By approximating the Adaptive AutoRegressive (AAR) model, the SRE time series was transformed into a multidimensional vector series regarded as a description of current users visiting patterns. Furthermore, a Support Vector Machine (SVM) classifier was applied to classify vector series and identify the attacks. The simulation results show that this approach not only can distinguish between normal traffic and DDoS attack traffic, but also is suitable to detect DDoS attack against the large scale network traffic, which does not arouse the sharp changes of the network traffic.