Can you describe him for me? A technique for semantic person search in video

From a law enforcement standpoint, the ability to search for a person matching a semantic description (i.e. 1.8m tall, red shirt, jeans) is highly desirable. While a significant research effort has focused on person re-detection (the task of identifying a previously observed individual in surveillance video), these techniques require descriptors to be built from existing image or video observations. As such, person re-detection techniques are not suited to situations where footage of the person of interest is not readily available, such as a witness reporting a recent crime. In this paper, we present a novel framework that is able to search for a person based on a semantic description. The proposed approach uses size and colour cues, and does not require a person detection routine to locate people in the scene, improving utility in crowded conditions. The proposed approach is demonstrated with a new database that will be made available to the research community, and we show that the proposed technique is able to correctly localise a person in a video based on a simple semantic description.

A database for person re-identification in multi-camera surveillance networks

Person re-identification involves recognising individuals in different locations across a network of cameras and is a challenging task due to a large number of varying factors such as pose (both subject and camera) and ambient lighting conditions. Existing databases do not adequately capture these variations, making evaluations of proposed techniques difficult. In this paper, we present a new challenging multi-camera surveillance database designed for the task of person re-identification. This database consists of 150 unscripted sequences of subjects travelling in a building environment though up to eight camera views, appearing from various angles and in varying illumination conditions. A flexible XML-based evaluation protocol is provided to allow a highly configurable evaluation setup, enabling a variety of scenarios relating to pose and lighting conditions to be evaluated. A baseline person re-identification system consisting of colour, height and texture models is demonstrated on this database.

Person re-identification using group information

After first observing a person, the task of person re-identification involves recognising an individual at different locations across a network of cameras at a later time. Traditionally, this task has been performed by first extracting appearance features of an individual and then matching these features to the previous observation. However, identifying an individual based solely on appearance can be ambiguous, particularly when people wear similar clothing (i.e. people dressed in uniforms in sporting and school settings). This task is made more difficult when the resolution of the input image is small as is typically the case in multi-camera networks. To circumvent these issues, we need to use other contextual cues. In this paper, we use "group" information as our contextual feature to aid in the re-identification of a person, which is heavily motivated by the fact that people generally move together as a collective group. To encode group context, we learn a linear mapping function to assign each person to a "role" or position within the group structure. We then combine the appearance and group context cues using a weighted summation. We demonstrate how this improves performance of person re-identification in a sports environment over appearance based-features.

Multi-shot person re-identification via relational stein divergence

Person re-identification is particularly challenging due to significant appearance changes across separate camera views. In order to re-identify people, a representative human signature should effectively handle differences in illumination, pose and camera parameters. While general appearance-based methods are modelled in Euclidean spaces, it has been argued that some applications in image and video analysis are better modelled via non-Euclidean manifold geometry. To this end, recent approaches represent images as covariance matrices, and interpret such matrices as points on Riemannian manifolds. As direct classification on such manifolds can be difficult, in this paper we propose to represent each manifold point as a vector of similarities to class representers, via a recently introduced form of Bregman matrix divergence known as the Stein divergence. This is followed by using a discriminative mapping of similarity vectors for final classification. The use of similarity vectors is in contrast to the traditional approach of embedding manifolds into tangent spaces, which can suffer from representing the manifold structure inaccurately. Comparative evaluations on benchmark ETHZ and iLIDS datasets for the person re-identification task show that the proposed approach obtains better performance than recent techniques such as Histogram Plus Epitome, Partial Least Squares, and Symmetry-Driven Accumulation of Local Features.

An efficient and robust system for multi-person event detection in real world indoor surveillance scenes

Due to the popularity of security cameras in public places, it is of interest to design an intelligent system that can efficiently detect events automatically. This paper proposes a novel algorithm for multi-person event detection. To ensure greater than real-time performance, features are extracted directly from compressed MPEG video. A novel histogram-based feature descriptor that captures the angles between extracted particle trajectories is proposed, which allows us to capture motion patterns of multi-person events in the video. To alleviate the need for fine-grained annotation, we propose the use of Labelled Latent Dirichlet Allocation, a “weakly supervised” method that allows the use of coarse temporal annotations which are much simpler to obtain. This novel system is able to run at approximately ten times real-time, while preserving state-of-theart detection performance for multi-person events on a 100-hour real-world surveillance dataset (TRECVid SED).

Data-Augmentation for Reducing Dataset Bias in Person Re-identification

In this paper we explore ways to address the issue of dataset bias in person re-identification by using data augmentation to increase the variability of the available datasets, and we introduce a novel data augmentation method for re-identification based on changing the image background. We show that use of data augmentation can improve the cross-dataset generalisation of convolutional network based re-identification systems, and that changing the image background yields further improvements.

Recurrent Convolutional Network for Video-based Person Re-Identification

In this paper we propose a novel recurrent neural networkarchitecture for video-based person re-identification.Given the video sequence of a person, features are extracted from each frame using a convolutional neural network that incorporates a recurrent final layer, which allows information to flow between time-steps. The features from all time steps are then combined using temporal pooling to give an overall appearance feature for the complete sequence. The convolutional network, recurrent layer, and temporal pooling layer, are jointly trained to act as a feature extractor for video-based re-identification using a Siamese network architecture.Our approach makes use of colour and optical flow information in order to capture appearance and motion information which is useful for video re-identification. Experiments are conduced on the iLIDS-VID and PRID-2011 datasets to show that this approach outperforms existing methods of video-based re-identification.

https://github.com/niallmcl/Recurrent-Convolutional-Video-ReID
Project Source Code

Person Re-Identification using Deep Convnets with Multi-task Learning

Person re-identification involves recognizing a person across non-overlapping camera views, with different pose, illumination, and camera characteristics. We propose to tackle this problem by training a deep convolutional network to represent a person’s appearance as a low-dimensional feature vector that is invariant to common appearance variations encountered in the re-identification problem. Specifically, a Siamese-network architecture is used to train a feature extraction network using pairs of similar and dissimilar images. We show that use of a novel multi-task learning objective is crucial for regularizing the network parameters in order to prevent over-fitting due to the small size the training dataset. We complement the verification task, which is at the heart of re-identification, by training the network to jointly perform verification, identification, and to recognise attributes related to the clothing and pose of the person in each image. Additionally, we show that our proposed approach performs well even in the challenging cross-dataset scenario, which may better reflect real-world expected performance. 

Determining operational measures from multi-camera surveillance systems using soft biometrics

CCTV and surveillance networks are increasingly being used for operational as well as security tasks. One emerging area of technology that lends itself to operational analytics is soft biometrics. Soft biometrics can be used to describe a person and detect them throughout a sparse multi-camera network. This enables them to be used to perform tasks such as determining the time taken to get from point to point, and the paths taken through an environment by detecting and matching people across disjoint views. However, in a busy environment where there are 100's if not 1000's of people such as an airport, attempting to monitor everyone is highly unrealistic. In this paper we propose an average soft biometric, that can be used to identity people who look distinct, and are thus suitable for monitoring through a large, sparse camera network. We demonstrate how an average soft biometric can be used to identify unique people to calculate operational measures such as the time taken to travel from point to point.

Identifying customer behaviour and dwell time using soft biometrics

In a commercial environment, it is advantageous to know how long it takes customers to move between different regions, how long they spend in each region, and where they are likely to go as they move from one location to another. Presently, these measures can only be determined manually, or through the use of hardware tags (i.e. RFID). Soft biometrics are characteristics that can be used to describe, but not uniquely identify an individual. They include traits such as height, weight, gender, hair, skin and clothing colour. Unlike traditional biometrics, soft biometrics can be acquired by surveillance cameras at range without any user cooperation. While these traits cannot provide robust authentication, they can be used to provide identification at long range, and aid in object tracking and detection in disjoint camera networks. In this chapter we propose using colour, height and luggage soft biometrics to determine operational statistics relating to how people move through a space. A novel average soft biometric is used to locate people who look distinct, and these people are then detected at various locations within a disjoint camera network to gradually obtain operational statistics

Searching for people using semantic soft biometric descriptions

It is not uncommon to hear a person of interest described by their height, build, and clothing (i.e. type and colour). These semantic descriptions are commonly used by people to describe others, as they are quick to communicate and easy to understand. However such queries are not easily utilised within intelligent video surveillance systems, as they are difficult to transform into a representation that can be utilised by computer vision algorithms. In this paper we propose a novel approach that transforms such a semantic query into an avatar in the form of a channel representation that is searchable within a video stream. We show how spatial, colour and prior information (person shape) can be incorporated into the channel representation to locate a target using a particle-filter like approach. We demonstrate state-of-the-art performance for locating a subject in video based on a description, achieving a relative performance improvement of 46.7% over the baseline. We also apply this approach to person re-detection, and show that the approach can be used to re-detect a person in a video steam without the use of person detection.

Communication-efficient decentralized sequential detection

The problem of decentralized sequential detection is studied in this thesis, where local sensors are memoryless, receive independent observations, and no feedback from the fusion center. In addition to traditional criteria of detection delay and error probability, we introduce a new constraint: the number of communications between local sensors and the fusion center. This metric is able to reflect both the cost of establishing communication links as well as overall energy consumption over time. A new formulation for communication-efficient decentralized sequential detection is proposed where the overall detection delay is minimized with constraints on both error probabilities and the communication cost. Two types of problems are investigated based on the communication-efficient formulation: decentralized hypothesis testing and decentralized change detection. In the former case, an asymptotically person-by-person optimum detection framework is developed, where the fusion center performs a sequential probability ratio test based on dependent observations. The proposed algorithm utilizes not only reported statistics from local sensors, but also the reporting times. The asymptotically relative efficiency of proposed algorithm with respect to the centralized strategy is expressed in closed form. When the probabilities of false alarm and missed detection are close to one another, a reduced-complexity algorithm is proposed based on a Poisson arrival approximation. In addition, decentralized change detection with a communication cost constraint is also investigated. A person-by-person optimum change detection algorithm is proposed, where transmissions of sensing reports are modeled as a Poisson process. The optimum threshold value is obtained through dynamic programming. An alternative method with a simpler fusion rule is also proposed, where the threshold values in the algorithm are determined by a combination of sequential detection analysis and constrained optimization. In both decentralized hypothesis testing and change detection problems, tradeoffs in parameter choices are investigated through Monte Carlo simulations.

Leveraging Multiple Features for Image Retrieval and Matching

The goal of image retrieval and matching is to find and locate object instances in images from a large-scale image database. While visual features are abundant, how to combine them to improve performance by individual features remains a challenging task. In this work, we focus on leveraging multiple features for accurate and efficient image retrieval and matching. We first propose two graph-based approaches to rerank initially retrieved images for generic image retrieval. In the graph, vertices are images while edges are similarities between image pairs. Our first approach employs a mixture Markov model based on a random walk model on multiple graphs to fuse graphs. We introduce a probabilistic model to compute the importance of each feature for graph fusion under a naive Bayesian formulation, which requires statistics of similarities from a manually labeled dataset containing irrelevant images. To reduce human labeling, we further propose a fully unsupervised reranking algorithm based on a submodular objective function that can be efficiently optimized by greedy algorithm. By maximizing an information gain term over the graph, our submodular function favors a subset of database images that are similar to query images and resemble each other. The function also exploits the rank relationships of images from multiple ranked lists obtained by different features. We then study a more well-defined application, person re-identification, where the database contains labeled images of human bodies captured by multiple cameras. Re-identifications from multiple cameras are regarded as related tasks to exploit shared information. We apply a novel multi-task learning algorithm using both low level features and attributes. A low rank attribute embedding is joint learned within the multi-task learning formulation to embed original binary attributes to a continuous attribute space, where incorrect and incomplete attributes are rectified and recovered. To locate objects in images, we design an object detector based on object proposals and deep convolutional neural networks (CNN) in view of the emergence of deep networks. We improve a Fast RCNN framework and investigate two new strategies to detect objects accurately and efficiently: scale-dependent pooling (SDP) and cascaded rejection classifiers (CRC). The SDP improves detection accuracy by exploiting appropriate convolutional features depending on the scale of input object proposals. The CRC effectively utilizes convolutional features and greatly eliminates negative proposals in a cascaded manner, while maintaining a high recall for true objects. The two strategies together improve the detection accuracy and reduce the computational cost.

Random projections on manifolds of symmetric positive definite matrices for image classification

Recent advances suggest that encoding images through Symmetric Positive Definite (SPD) matrices and then interpreting such matrices as points on Riemannian manifolds can lead to increased classification performance. Taking into account manifold geometry is typically done via (1) embedding the manifolds in tangent spaces, or (2) embedding into Reproducing Kernel Hilbert Spaces (RKHS). While embedding into tangent spaces allows the use of existing Euclidean-based learning algorithms, manifold shape is only approximated which can cause loss of discriminatory information. The RKHS approach retains more of the manifold structure, but may require non-trivial effort to kernelise Euclidean-based learning algorithms. In contrast to the above approaches, in this paper we offer a novel solution that allows SPD matrices to be used with unmodified Euclidean-based learning algorithms, with the true manifold shape well-preserved. Specifically, we propose to project SPD matrices using a set of random projection hyperplanes over RKHS into a random projection space, which leads to representing each matrix as a vector of projection coefficients. Experiments on face recognition, person re-identification and texture classification show that the proposed approach outperforms several recent methods, such as Tensor Sparse Coding, Histogram Plus Epitome, Riemannian Locality Preserving Projection and Relational Divergence Classification.