Search

Dual-Band Infrared (DBIR) offers the advantage of combining Mid-Wave Infrared (MWIR) and Long-Wave Infrared (LWIR) within a single field-of-view (FoV). This provides additional information for each spectral band. DBIR camera systems find applications in both military and civilian contexts. This work introduces a novel labeled DBIR dataset that includes civilian vehicles, aircraft, birds, and people. The dataset is designed for utilization in object detection and tracking algorithms. It comprises 233 objects with tracks spanning up to 1,300 frames, encompassing images in both MW and LW. This research reviews pertinent literature related to object detection, object detection in the infrared spectrum, and data fusion. Two sets of experiments were conducted using this DBIR dataset: Motion Detection and CNNbased object detection. For motion detection, a parallel implementation of the Visual Background Extractor (ViBe) was developed, employing ConnectedComponents analysis to generate bounding boxes. To assess these bounding boxes, Intersection-over-Union (IoU) calculations were performed. The results demonstrate that DBIR enhances the IoU of bounding boxes in 6.11% of cases within sequences where the camera’s field of view remains stationary. A size analysis reveals ViBe’s effectiveness in detecting small and dim objects within this dataset. A subsequent experiment employed You Only Look Once (YOLO) versions 4 and 7 to conduct inference on this dataset, following image preprocessing. The inference models were trained using visible spectrum MS COCO data. The findings confirm that YOLOv4/7 effectively detect objects within the infrared spectrum in this dataset. An assessment of these CNNs’ performance relative to the size of the detected object highlights the significance of object size in detection capabilities. Notably, DBIR substantially enhances detection capabilities in both YOLOv4 and YOLOv7; however, in the latter case, the number of False Positive detections increases. Consequently, while DBIR improves the recall of YOLOv4/7, the introduction of DBIR information reduces the precision of YOLOv7. This study also demonstrates the complementary nature of ViBe and YOLO in their detection capabilities based on object size in this data set. Though this is known prior art, an approach using these two approaches in a hybridized configuration is discussed. ViBe excels in detecting small, distant objects, while YOLO excels in detecting larger, closer objects. The research underscores that DBIR offers multiple advantages over MW or LW alone in modern computer vision algorithms, warranting further research investment.

The main objective of this thesis is to propose a localization method for a UAV using various computer vision and machine learning techniques. It plays a major role in planning the strategy for the flight, and acts as a navigational contingency method, in event of a GPS failure. The implementation of the algorithms employs high processing capabilities of the graphics processing unit, making it more efficient. The method involves the working of various neural networks, working in synergy to perform the localization. This thesis is a part of a collaborative project between The University of North Texas, Denton, USA, and the University of Windsor, Ontario, Canada. The localization has been divided into three phases namely object detection, recognition, and location estimation. Object detection and position estimation were discussed in this thesis while giving a brief understanding of the recognition. Further, future strategies to aid the UAV to complete the mission, in case of an eventuality, like the introduction of an EDGE server and wireless charging methods, was also given a brief introduction.

Current methods in computer vision and object detection rely heavily on neural networks and deep learning. This active area of research is used in applications such as autonomous driving, aerial imaging, defense and surveillance. State-of-the-art object detection methods rely on rectangular shaped, horizontal/vertical bounding boxes drawn over an object to accurately localize its position. Such orthogonal bounding boxes ignore object pose, resulting in reduced object localization, and limiting downstream tasks such as object understanding and tracking. To overcome these limitations, this research presents object detection improvements that aid tighter and more precise detections. In particular, we modify the object detection anchor box definition to firstly include rotations along with height and width and secondly to allow arbitrary four corner point shapes. Further, the introduction of new anchor boxes gives the model additional freedom to model objects which are centered about a 45-degree axis of rotation. The resulting network allows minimum compromises in speed and reliability while providing more accurate localization. We present results on the DOTA dataset, showing the value of the flexible object boundaries, especially with rotated and non-rectangular objects.