\prefacesection{Abstract} Multi-object tracking (MOT) is a key building block in video analytics and finds extensive use in surveillance, search and rescue, and autonomous driving applications. Object detection, a crucial stage in MOT, dominates in the overall tracking inference time due to its reliance on Deep Neural Networks (DNNs). Despite the superior performance of cutting-edge object detectors, their extensive computational demands limit their real-time application on embedded devices that possess constrained processing capabilities. Hence, we aim to reduce the computational burdens of object detection while maintaining tracking performance. As the first approach, we adapt frame resolutions to reduce computational complexity. During inference, frame resolutions can be tuned according to the complexity of visual scenes. We present DeepScale, a model-agnostic frame resolution selection approach that operates on top of existing fully convolutional network-based trackers. By analyzing the effect of frame resolution on detection performance, DeepScale strikes good trade-offs between detection accuracy and processing speed by adapting frame resolutions on-the-fly. Our second approach focuses on enhancing the efficiency of a tracker by model adaptation. We introduce AttTrack to expedite tracking by interleaving the execution of object detectors of different model sizes in inference. A sophisticated network (teacher) runs for keyframes only while, for non-keyframe, knowledge is transferred from the teacher to a smaller network (student) to improve the latter’s performance. Our third contribution involves exploiting temporal-spatial redundancies to enable real-time multi-camera tracking. We propose the MVSparse pipeline which consists of a central processing unit that aggregates information from multiple cameras (on an edge server or in the cloud) and distributed lightweight Reinforcement Learning (RL) agents running on individual cameras that predict the informative blocks in the current frame based on past frames on the same camera and detection results from other cameras. / Thesis / Doctor of Science (PhD)

In summary the tilting mechanism helps to explain the overall flow structure and the stability of the leading edge vortex.; The leading edge vortex has been identified as the most critical flow structure for producing lift in flapping wing flight. Its stability depends on the transport of the entrained vorticity into the wake via spanwise flow. This study proposes a hypothesis for the generation and enhancement of spanwise flow based on the chordwise vorticity that results from the tilting of the leading edge vortex and trailing edge vortex. We investigate this phenomenon using dynamically scaled robotic model wings. Two different wing shapes, one rectangular and one based on Drosophila melanogaster (fruit fly), are submerged in a tank of mineral oil and driven in a flapping motion. Two separate kinematics, one of constant angular velocity and one of sinusoidal angular velocity are implemented. In order to visualize the flow structure, a novel three dimensional particle image velocimetry system is utilized. From the three dimensional information obtained the chordwise vorticity resulting from the vortex tilting is shown using isosurfaces and planar slices in the wake of the wing. It is observed that the largest spanwise flow is located in the area between the chordwise vorticity of the leading edge vortex and the chordwise vorticity of the trailing edge vortex, supporting the hypothesis that the vortex tilting enhances the spanwise flow. Additionally the LEV on the rectangular wing is found to detach at about 80% span as opposed to 60% span for the elliptical wing. Also, two distinct regions of spanwise flow, one at the base and one at the tip, are observed at the beginning of the sinusoidal kinematic, and as the velocity of the wing increases these two regions unionize into one. Lastly, the general distribution of vorticity around each wing is found to be nearly the same, indicating that different wing shapes do not greatly affect the distribution of vorticity nor stability mechanisms in flapping flight.

Dating back to the 1930's, Ramsey theory still intrigues many who study combinatorics. Roughly put, it makes the profound assertion that complete disorder is impossible. One view of this problem is in edge-colorings of complete graphs. For forbidden graphs H1,...,Hk and a graph G, we write G "arrows" (H1,...,Hk) if every k-edge-coloring of G contains a monochromatic copy of Hi in color i for some i=1,2,...,k. If c is a (red, blue)-edge-coloring of G, we say c is a bad coloring if G contains no red K3or blue K1,t under c. A graph G is (H1,...,Hk)-Ramsey-minimal if G arrows (H1,...,Hk) but no proper subgraph of G has this property. Given a family F of graphs, we say that a graph G is F-saturated if no member of F is a subgraph of G, but for any edge xy not in E(G), G + xy contains a member of F as a subgraph. Letting Rmin(K3, K1,t) be the family of (K3,K1,t)-Ramsey minimal graphs, we study the saturation number, denoted sat(n,Rmin(K3,K1,t)), which is the minimum number of edges among all Rmin(K3,K1,t)-saturated graphs on n vertices. We believe the methods and constructions developed in this thesis will be useful in studying the saturation numbers of (K4,K1,t)-saturated graphs.

Managed forest ecosystems contribute to conservation of wildlife, and generally yield heterogeneous landscapes with patches that support different sessile organisms. Edges or boundaries between adjacent forest patches are often ecologically distinct from patch interiors and can have direct influences on community dynamics and ecosystem functioning near them. To quantify effects of edge and adjacent habitat conditions on avian metrics, I used a hierarchical multi-species occupancy model that considered individual species resource requirements to estimate community occupancy patterns, and used artificial nest surveys to model daily nest survival (DNS) using a Bayesian framework. Results indicated that adjacent forest conditions influenced bird population dynamics in focal forest stands, provided little evidence of an edge effect on avian community response patterns, and showed highest DNS in newly established forest stands. My results highlight the importance of considering type and spatial arrangement of different habitat patches for habitat planning operations on managed forest landscapes.

One knowledge gap hindering prairie restoration is uncertainty about when a restored prairie communities sufficiently resemble remnant prairie. I surveyed plant communities in remnant prairies, prairies > 5 years post-restoration, and prairies ≤ 5 years post-restoration in Mississippi. Remnants had the greater species richness. Restored prairies had less cover of woody plants and forbs but greatest non-natives. Restored prairies were not similar to remnant prairies (similarity index = 28.9 - 25.9%), primarily because restored prairies had fewer prairie forbs. Thus, restoration may take decades. Transplanting locallyapted prairie forbs into restored prairies may accelerate restoration, but this has not been evaluated adequately. I transplanted a prairie forb (Liatris pycnostachya) into prepared beds, oldields, and restored prairies. Prepared beds had greater growth and seed production, but survival and flowering was high in oldields and restored prairies. Augmenting restored prairies with locallyapted forbs has promise for accelerating prairie restoration.

No description available.

No description available.

No description available.

No description available.

No description available.