<p>Deep learning solutions to computer vision tasks have revolutionized many industries in recent years, but embedded systems have too many restrictions to take advantage of current state-of-the-art configurations. Typical embedded processor hardware configurations must meet very low power and memory constraints to maintain small and lightweight packaging, and the architectures of the current best deep learning models are too computationally intensive for these hardware configurations. Current research shows that convolutional neural networks (CNNs) can be deployed with a few architectural modifications on Field-Programmable Gate Arrays (FPGAs) resulting in minimal loss of accuracy, similar or decreased processing speeds, and lower power consumption when compared to general-purpose Central Processing Units (CPUs) and Graphics Processing Units (GPUs). This research contributes further to these findings with the FPGA implementation of a YOLOv4 object detection model that was developed with the use of transfer learning. The transfer-learned model uses the weights of a model pre-trained on the MS-COCO dataset as a starting point then fine-tunes only the output layers for detection on more specific objects of five classes. The model architecture was then modified slightly for compatibility with the FPGA hardware using techniques such as weight quantization and replacing unsupported activation layer types. The model was deployed on three different hardware setups (CPU, GPU, FPGA) for inference on a test set of images. It was found that the FPGA was able to achieve real-time inference speeds of 33.77 frames-per-second, a speedup of 7.74 frames-per-second when compared to GPU deployment. The model also consumed 96% less power than a GPU configuration with only approximately 4% average loss in accuracy across all 5 classes. The results are even more striking when compared to CPU deployment, with 131.7-times speedup in inference throughput. CPUs have long since been outperformed by GPUs for deep learning applications but are used in most embedded systems. These results further illustrate the advantages of FPGAs for deep learning inference on embedded systems even when transfer learning is used for an efficient end-to-end deployment process. This work advances current state-of-the-art with the implementation of a YOLOv4 object detection model developed with transfer learning for FPGA deployment.</p>

Automatic Passenger Counting (APC) systems are some of the many Internet-Of-Things (IoT) applications and have been increasingly adopted by public transportation companies in recent years. APCs provide valuable data that can be used to give an real time passenger count, which can be a convenient service and allow customers to plan their travels accordingly. The provided data is also valuable for resource streamlining and planning, which potentially increases revenues for the public transportation companies. This thesis briefly studies and evaluates different APC technologies, highlights the advantages and disadvantages of these, and presents an Edge-prototype based on Computer Vision and Object Detection. The presented APC was tested in a lab environment and with recordings of people walking in and out of a designated area in the lab. Test results from the lab environment show that the presented low-cost APC efficiently detects passengers with an accuracy of 98.6% on pre-recorded videos. The APC was also tested in real time and the results show that the low-cost APC only achieved an accuracy of 66.7%. This work has laid the ground for further development and testing in a public transport environment.

Indianapolis / E-scooters are ubiquitous and their number keeps escalating, increasing their interactions with other vehicles on the road. E-scooter riders have an atypical behavior that varies enormously from other vulnerable road users, creating new challenges for vehicle active safety systems and automated driving functionalities. The detection of e-scooter riders by other vehicles is the first step in taking care of the risks. This research presents a novel vision-based system to differentiate between e-scooter riders and regular pedestrians and a benchmark dataset for e-scooter riders in natural environments. An efficient system pipeline built using two existing state-of-the-art convolutional neural networks (CNN), You Only Look Once (YOLOv3) and MobileNetV2, performs detection of these vulnerable e-scooter riders.

Sensors, as a crucial part of autonomous driving, are primarily used for perceiving the environment. The recent deep learning development of different sensors has demonstrated the ability of machines recognizing and understanding their surroundings. Automotive radar, as a primary sensor for self-driving vehicles, is well-known for its robustness against variable lighting and weather conditions. Compared with camera-based deep learning development, Object detection using automotive radars has not been explored to its full extent. This can be attributed to the lack of public radar datasets. In this thesis, we collect a novel radar dataset that contains radar data in the form of Range-Azimuth-Doppler tensors along with the bounding boxes on the tensor for dynamic road users, category labels, and 2D bounding boxes on the Cartesian Bird-EyeView range map. To build the dataset, we propose an instance-wise auto-annotation algorithm. Furthermore, a novel Range-Azimuth-Doppler based multi-class object detection deep learning model is proposed. The algorithm is a one-stage anchor-based detector that generates both 3D bounding boxes and 2D bounding boxes on Range-AzimuthDoppler and Cartesian domains, respectively. Our proposed algorithm achieves 56.3% AP with IOU of 0.3 on 3D bounding box predictions, and 51.6% with IOU of 0.5 on 2D bounding box predictions. Our dataset and the code can be found at https://github.com/ZhangAoCanada/RADDet.git.

Master’s thesis deals with object pose estimation using monocular camera. As an object is considered every rigid, shape fixed entity with strong edges, ideally textureless. Object position in this work is represented by transformation matrix, which describes object translation and rotation towards world coordinate system. First chapter is dedicated to explanation of theory of geometric transformations and intrinsic and extrinsic parameters of camera. This chapter also describes detection algorithm Chamfer Matching, which is used in this work. Second chapter describes all development tools used in this work. Third, fourth and fifth chapter are dedicated to practical realization of this works goal and achieved results. Last chapter describes created application, that realizes known object pose estimation in scene.

Tato práce se zabývá detekcí chodců pomocí konvolučních neuronových sítí z pohledu autonomního vozidla. A to zejména jejich otestováním ve smyslu nalezení vhodné praxe tvorby datasetu pro machine learning modely. V práci bylo natrénováno celkem deset machine learning modelů meta architektur Faster R-CNN s ResNet 101 jako feature extraktorem a SSDLite s feature extraktorem MobileNet_v2. Tyto modely byly natrénovány na datasetech o různých velikostech. Nejlépší výsledky byly dosaženy na datasetu o velikosti 5000 snímků. Kromě těchto modelů byl vytvořen nový dataset zaměřující se na chodce v noci. Dále byla vytvořena knihovna Python funkcí pro práci s datasety a automatickou tvorbu datasetu.

This master's thesis deals with automatic detection and tracking of a soccer ball in sports videos. Based on the introduced techniques focusing on tracking of small objects in high-resolution videos, effective convolutional neural networks are designed and used by a modified version of tracking algorithm SORT for automatic object detection. A set of experiments with the processing of images in different resolutions and with various frequencies of detection extraction is carried out in order to examine the trade-off between processing speed and tracking accuracy. The obtained results of experiments are presented and used to form proposals for future work, which could lead to improvements in tracking accuracy while maintaining reasonable processing speed.

Tato diplomová práce pojednává o dvou algoritmech pro dolování z proudu dat - Very Fast Decision Tree (VFDT) a Concept-adapting Very Fast Decision Tree (CVFDT). Je vysvětlen princip klasifikace rozhodovacím stromem. Je popsána základní myšlenka konstrukce stromu Hoeffding Tree, který je základem pro algoritmy VFDT a CVFDT. Tyto algoritmy jsou poté rozebrány detailněji. Dále se tato práce zabývá návrhem algoritmu Genetického Programování (GP), který je použit pro vytváření klasifikátoru obrazových dat. Vytvořený klasifikátor je použit jako alternativní způsob klasifikace objektů v obraze ve frameworku Viola-Jones. V práci je rozebrána implementace algoritmů, které jsou implementovány v jazyce Java. Algoritmus GP je integrován do knihovny “Image Processing Extension” programu RapidMiner. Algoritmy VFDT a CVFDT jsou testovány na syntetických a reálných textových datech. Algoritmus GP je testován na klasifikaci obrazových dat a následně vytvořený klasifikátor je otestován na detekci obličejů v obraze.

This thesis deals with the image object detection using Haar--like features and AdaBoost algorithm. The text describes methods how to train and test an object detector. The main contributon of this thesis consists in creation image object detector in Java programming language. Created algorithms were integrated as plugin into the RapidMiner tool, which is widely used and known worldwide as tool for data mining. The thesis contains the instructions for created operators and few exaples for executing in RapidMiner tool. The functionality of image object detector was demonstrated on selected medical images.

This thesis deals with the image object detection using Contour cues. The text describes methods how to train and test object detector. The main contribution of this thesis consists in creation Feature extractor for creation object detector in Java programming. The functionality of object detector was demonstrated on medical images.