Distributed target tracking (DTT) is desirable in wireless camera networks to achieve scalability and robustness to node or link failures. DTT estimates the target state via information exchange and fusion among cameras. This thesis proposes new DTT algorithms to handle five major challenges of DTT in wireless camera networks, namely non-linearity in the camera measurement model, temporary lack of measurements (benightedness) due to limited field of view, redundant information in the network, limited connectivity of the network due to limited communication ranges and asynchronous information caused by varying and unknown frame processing delays. The algorithms consist of two phases, namely estimation and fusion. In the estimation phase, the cameras process their captured frames, detect the target, and estimate the target state (location and velocity) and its uncertainty using the Extended Information Filter (EIF) that handles non-linearity. In the fusion phase, the cameras exchange their local target information with their communicative neighbours and fuse the information. The contributions of this thesis are as follows. The target states estimated by the EIFs undergo weighted fusion. The weights are chosen based on the estimated uncertainty (error covariance) and the number of nodes with redundant information such that the information of benighted nodes and the redundant information get lower weights. At each time step, only the cameras having the view of the target and the cameras that might have the view of the target in the next time step participate in the fusion (tracking). This reduces the energy consumption of the network. The algorithm selects the cameras dynamically by using a threshold on their shortest distances (in the communication graph) from the cameras having the view of the target. Before fusion, each camera predicts the target information of other cameras to temporally align its information with the (asynchronous) information received from other cameras. The algorithm predicts the target state using the latest estimated velocity of the target. The experimental results show that the proposed algorithms achieve higher tracking accuracy than the state of the art under the five DTT challenges.

With the proliferation of cameras and advanced video analytics, situation awareness applications that automatically generate actionable knowledge from live camera streams has become an important class of applications in various domains including surveillance, marketing, sports, health care, and traffic monitoring. However, despite the wide range of use cases, developing those applications on large-scale camera networks is extremely challenging because it involves both compute- and data-intensive workloads, has latency-sensitive quality of service requirement, and deals with inherent dynamism (e.g., number of faces detected in a certain area) from the real world. To support developing large-scale situation awareness applications, this dissertation presents a distributed framework that makes two key contributions: 1) it provides a programming model that ensures scalability of applications and 2) it supports low-latency computation and dynamic workload handling through opportunistic event processing and workload distribution over different locations and network hierarchy. To provide a scalable programming model, two programming abstractions for different levels of application logic are proposed: the first abstraction at the level of real-time target detection and tracking, and the second abstraction for answering spatio-temporal queries at a higher level. The first programming abstraction, Target Container (TC), elevates target as a first-class citizen, allowing domain experts to simply provide handlers for detection, tracking, and comparison of targets. With those handlers, TC runtime system performs priority-aware scheduling to ensure real-time tracking of important targets when resources are not enough to track all targets. The second abstraction, Spatio-temporal Analysis (STA) supports applications to answer queries related to space, time, and occupants using a global state transition table and probabilistic events. To ensure scalability, STA supports bounded communication overhead of state update by providing tuning parameters for the information propagation among distributed workers. The second part of this work explores two optimization strategies that reduce latency for stream processing and handle dynamic workload. The first strategy, an opportunistic event processing mechanism, performs event processing on predicted locations to provide just-in-time situational information to mobile users. Since location prediction algorithms are inherently inaccurate, the system selects multiple regions using a greedy algorithm to provide highly meaningful information at the given amount of computing resources. The second strategy is to distribute application workload over computing resources that are placed at different locations and various levels of network hierarchy. To support this strategy, the framework provides hierarchical communication primitives and a decentralized resource discovery protocol that allow scalable and highly adaptive load balancing over space and time.

Distributed Smart Camera (DSC) networks are power constrained real-time distributed embedded systems that perform computer vision using multiple cameras. Providing data aggregation techniques that is criti-cal for running complex image processing algorithms on DSCs is a challenging task due to complexity of video and image data. Providing highly desirable SQL APIs for sophisticated query processing in DSC networks is also challenging for similar reasons. Research on DSCs to date have not addressed the above two problems. In this thesis, we develop a novel SOA based middleware framework on a DSC network that uses Distributed OSGi to expose DSC network services as web services. We also develop a novel web service composition scheme that aid in data aggregation and a SQL query interface for DSC net-works that allow sophisticated query processing. We validate our service orchestration concept for data aggregation by providing query primitive for face detection in smart camera network.

Η παρούσα διπλωματική εργασία μελετά την ανίχνευση και παρακολούθηση της κίνησης των ανθρώπων μέσα από δίκτυα καμερών. Σκοπός της παρούσας εργασίας είναι η υλοποίηση ενός συστήματος ανίχνευσης , παρακολούθησης εκ νέου ταυτοποίησης των ανθρώπων που διέρχονται μέσα από ένα δίκτυο καμερών καθώς και να προτείνει ένα μοντέλο για την κατανόηση της τοπολογίας του δικτύου των καμερών. Το κύριο πρόβλημα υποδιαιρείται σε τρία επιμέρους υπό – προβλήματα. Το πρώτο αφορά την ανίχνευση κίνησης. Το δεύτερο την παρακολούθηση των ανθρώπων και τέλος το τρίτο αφορά την αντιστοίχηση τους μεταξύ των καμερών. Σαν αποτέλεσμα στο τέλος έχουμε για κάθε άνθρωπο το μονοπάτι που διέγραψε μέσα στο δίκτυο. Η Ανίχνευση κίνησης υλοποιείται με αφαίρεση φόντου. Η παρακολούθηση υλοποιείται με δύο χαρακτηριστικά, αυτά του κέντρου μάζας και του χρωματικού ιστογράμματος. Η τοπολογία του δικτύου ανακαλύπτεται με ένα μοντέλο που καταγράφει σημεία εισόδου και εξόδου συσχετισμένα με την αντίστοιχη κάμερα από την οποία εισήλθαν ή στην οποία εξήλθαν αντίστοιχα οι άνθρωποι. Κατόπιν γίνεται αντιστοίχηση των σημείων αυτών στις κρίσιμες περιοχές της κάθε κάμερας και η πλειοψηφία των συσχετίσεων τους ορίζει την επικοινωνούσα , για αυτές τις περιοχές , κάμερα. Τέλος γίνεται η αντιστοίχηση των διαδρομών μεταξύ καμερών με έλεγχο χώρο-χρονικών χαρακτηριστικών και χαρακτηριστικών εμφάνισης. Το σύστημα υλοποιήθηκε σε Matlab και έτρεξε σε Intel i7 με συχνότητα 2.93 Ghz και 8GB μνήμης ram. Οι αλγόριθμοι λειτούργησαν ικανοποιητικά με πολύ καλά αποτελέσματα, και μπορούν να περάσουν ως είσοδοι σε πληθώρα εφαρμογών υψηλοτέρου επιπέδου που έχουν ως σκοπό την αναγνώριση της ανθρώπινης δραστηριότητας και την κατανόηση συμπεριφοράς. / This thesis deals with the detection and motion tracking through camera networks. Its purpose is to implement a system for monitoring human movement and perform re-identification in camera networks. It also proposes a model for discovering the topology of cameras network. The main problem is divided into three sub – problems. The first one deals with motion detection , the second one tracks every human located in the plane, and finally the third one has to do with the re-identification between the cameras. As a result we find and identify all human’s paths traced in the network. At first we start with detection that involves also background subtraction. The background is recovered in a dynamic way at every frame and involves median selection. Tracking is accomplished using two features, the centroid and the color histogram. Network topology is discovered from a model which reports entry and exit points associated with the corresponding camera. The system is implemented in Matlab and runs on Intel i7 with frequency 2.93 Ghz and 8GB of ram. The algorithms perform well producing very good results, and can be fed as inputs to a variety of applications that deal with problems related to higher level recognition of human activity and behavior understanding.

<div>Numerous organizations around the world deploy sensor networks, especially visual sensor networks for various applications like monitoring traffic, security, and emergencies. With advances in computer vision technology, the potential application of these sensor networks has expanded. This has led to an increase in demand for deployment of large scale sensor networks.</div><div>Sensors in a large network have differences in location, position, hardware, etc. These differences lead to varying usefulness as they provide different quality of information. As an example, consider the cameras deployed by the Department of Transportation (DOT). We want to know whether the same traffic cameras could be used for monitoring the damage by a hurricane.</div><div>Presently, significant manual effort is required to identify useful sensors for different applications. There does not exist an automated system which determines the usefulness of the sensors based on the application. Previous methods on visual sensor networks focus on finding the dependability of sensors based on only the infrastructural and system issues like network congestion, battery failures, hardware failures, etc. These methods do not consider the quality of information from the sensor network. In this paper, we present an automated system which identifies the most useful sensors in a network for a given application. We evaluate our system on 2,500 real-time live sensors from four cities for traffic monitoring and people counting applications. We compare the result of our automated system with the manual score for each camera.</div><div>The results suggest that the proposed system reliably finds useful sensors and it output matches the manual scoring system. It also shows that a camera network deployed for a certain application can also be useful for another application.</div>