In recent years, various high-performance networks have been proposed for single-image object detection. An obvious choice is to design a video detection network based on state-of-the-art single-image detectors. However, video object detection is still challenging due to the lower quality of individual frames in a video, and hence the need to include temporal information for high-quality detection results. In this thesis, we design a novel interleaved architecture combining a 2D convolutional network and a 3D temporal network. We utilize Yolov3 as the base detector. To explore inter-frame information, we propose feature aggregation based on a temporal network. Our temporal network utilizes Appearance-preserving 3D convolution (AP3D) for extracting aligned features in the temporal dimension. Our multi-scale detector and multi-scale temporal network communicate at each scale and also across scales. The number of inputs of our temporal network can be either 4, 8, or 16 frames in this thesis and correspondingly we name our temporal network TemporalNet-4, TemporalNet-8 and TemporalNet-16. Our approach achieves 77.1\% mAP (mean Average Precision) on ImageNet VID 2017 dataset with TemporalNet-4, where TemporalNet-16 achieves 80.9\% mAP which is a competitive result on this video object detection benchmark. Our network is also real-time with a running time of 35ms/frame.

Cílem této dizertační prace je vylepšit existující detektory objektů pomocí sdílení informace a výpočtů mezi blízkými pozicemi v obraze. Navrhuje dvě metody, které jsou založené na Waldově sekvenčním testu poměrem pravděpodobností a algoritmu WaldBoost. První z nich, Early non-Maxima Suppression , přesunuje rozhodování o potlačení nemaximálních pozic ze závěrečné fáze do fáze vyhodnocování detektoru, čímž zamezuje zbytečným výpočtům detektoru v nemaximálních pozicích. Metoda neighborhood suppression doplňuje existující detektory o schopnost zavrhnout okolní pozice v obraze. Navržené metody je možné aplikovat na širokou škálu detektorů. Vyhodnocení obou metod dokazují jejich výrazně vyšší efektivitu v porovnání s detektory, které vyhodnocují jednotlivé pozice obrazu zvlášť. Dizertace navíc prezentuje výsledky rozsáhlých experimentů, jejichž cílem bylo vyhodnotit vlastnosti běžných obrazových příznaků v několika detekčních úlohách a situacích.

Detekce objektů v počítačovém vidění je složítá úloha. Velmi populární a rozšířená metoda pro detekci je využití statistických klasifikátorů a skenovacích oken. Pro učení kalsifikátorů se často používá algoritmus AdaBoost (nebo jeho modifikace), protože dosahuje vysoké úspěšnosti detekce, nízkého počtu chybných detekcí a je vhodný pro detekci v reálném čase. Implementaci detekce objektů je možné provést různými způsoby a lze využít vlastnosti konkrétní architektury, pro urychlení detekce. Pro akceleraci je možné využít grafické procesory, vícejádrové architektury, SIMD instrukce, nebo programovatelný hardware. Tato práce představuje metodu optimalizace, která vylepšuje výkon detekce objektů s ohledem na cenovou funkci zadanou uživatelem. Metoda rozděluje předem natrénovaný klasifikátor do několika různých implementací, tak aby celková cena klasifikace byla minimalizována. Metoda je verifikována na základním experimentu, kdy je klasifikátor rozdělen do předzpracovací jednotku v FPGA a do jednotky ve standardním PC.

No description available.

Background: With the emerging application of autonomous vehicles in the automotive industry, several efforts have been made for the complete adoption of autonomous vehicles. One of the several problems in creating autonomous technology is the detection of water puddles, which can cause damages to internal components and the vehicle to lose control. This thesis focuses on the detection of water puddles on-road and off-road conditions with the use of Deep Learning models. Objectives: The thesis focuses on finding suitable Deep Learning algorithms for detecting the water puddles, and then an experiment is performed with the chosen algorithms. The algorithms are then compared with each other based on the performance evaluation of the trained models. Methods: The study uses a literature review to find the appropriate Deep Learning algorithms to answer the first research question, followed by conducting an experiment to compare and evaluate the selected algorithms. Metrics used to compare the algorithm include accuracy, precision, recall, f1 score, training time, and detection speed. Results: The Literature Review indicated Faster R-CNN and SSD are suitable algorithms for object detection applications. The experimental results indicated that on the basis of accuracy, recall, and f1 score, the Faster R-CNN is a better performing algorithm. But on the basis of precision, training time, and detection speed, the SSD is a faster performing algorithm. Conclusions: After carefully analyzing the results, Faster R-CNN is preferred for its better performance due to the fact that in a real-life scenario which the thesis aims at, the models to correctly predict the water puddles is key

Computer vision, a prominent subfield of artificial intelligence, has gained widespread util-ization in diverse domains such as surveillance, security, and robotics. This research en-deavors to develop an semi-automated defect detection system serving as a quality controlassurance mechanism for Nolato MediTor, a manufacturing company within the medicaldevice industries engaged in the production of anesthesia breathing bags. The primary fo-cus of this study revolves around the detection of a specific defect, namely, holes. Withinthe context of Nolato MediTor, prioritizing recall (sensitivity) assumes utmost signific-ance as it entails favoring the rejection of functional breathing bags over the inadvertentacceptance of defective ones. The proposed system encompasses a robust metallic standfacilitating precise positioning for three distinct camera angles, accompanied by a XiaomiRedmi Note 11 Pro phone and a software component, designed to process incoming imagefolders representing a complete view of a breathing bag from multiple angles. Subsequently,these images undergo analysis using the learned weights derived from the implementedMask R-CNN model, enabling a cohesive assessment of the breathing bag. The system’sperformance was rigorously evaluated, and the best-performing weights demonstrated aremarkable recall rate of 0.995 for the first test set, exceeding the desired recall thresholdof 95%. Similarly, for the second test set, the recall rate achieved an impressive value of0.949, narrowly missing the 95% threshold by a marginal 0.001. Furthermore, the com-putational efficiency, quantified as the processing time per breathing bag, on average, thelongest duration recorded amounted to approximately 10.151 seconds, with the poten-tial for further enhancement by employing a higher standard GPU. This study serves as aproof of concept, demonstrating the feasibility of achieving semi-automated quality controlutilizing CNN. The implemented system represents a promising prototype with potentialscalability for improved operational conditions and expanded defect coverage, thus pavingthe way towards a fully automated quality control within large-scale industries.

With the development of small satellites it has become easier and cheaper to deploy satellites for earth observation from space. While optical sensors capture high-resolution data, this data is traditionally sent to earth for analysis which puts a high constraint on the data link and increases the time for making data based decisions. This thesis explores the possibilities of deploying an AI model in small satellites for detecting objects in satellite imagery and therefore reduce the amount of data that needs to be transmitted. The neural network model YOLOv8 was trained on the xView and DIOR dataset and evaluated in a hardware restricted execution environment. The model achieved a mAP50 of 0.66 and could process satellite images at a speed of 309m2/s.

Perceptual categorization involves integrating bottom-up sensory information with top-down knowledge which is based on prior experience. Bottom-up information comes from the external world and visual saliency is a type of bottom-up information that is calculated on the differences between the visual characteristics of adjacent spatial locations. There is currently a related debate in municipal law enforcement communities about which are more ‘visible’: white police cars or black and white police cars. Municipalities do not want police cars to be hit by motorists and they also want police cars to be seen in order to promote a public presence. The present study used three behavioral experiments to investigate the effects of visual saliency on object detection and categorization. Importantly, the results indicated that so-called ‘object detection’ is not a valid construct. Rather than identifying objectness or objecthood prior to categorization, object categorization is an obligatory process, and object detection is a postcategorization decision with higher salience objects being categorized easier than lower salience objects. An additional experiment was conducted to examine the features that constitute a police car. Based on salience alone, black and white police cars were better categorized than white police cars and light bars were slightly more important police car defining components than markings.

The objective of the thesis is to explore an approach of classifying and localizing different objects from driving-scene images using YOLOv4 algorithm trained on custom dataset.  YOLOv4, a one-stage object detection algorithm, aims to have better accuracy and speed. The deep learning (convolutional) network based classification model was trained and validated on a subject of SODA10M dataset annotated with six different classes of objects (Car, Cyclist, Truck, Bus, Pedestrian, and Tricycle), which are the most seen objects on the road. Another model based on YOLOv3 (the previous version of YOLOv4) will be trained on the same dataset and the performance will be compared with the YOLOv4 model. Both algorithms are fast but have difficulty detecting some objects, especially the small objects. Larger quantities of properly annotated training data can improve the algorithm's performance accuracy.

The perception system is the key for autonomous vehicles to sense and understand the surrounding environment. As the cheapest and most mature sensor, monocular cameras create a rich and accurate visual representation of the world. The objective of this thesis is to investigate if camera-based deep learning models with transfer learning technique can achieve 2D object detection, License Plate Detection and Recognition (LPDR), and highway lane detection in real time. The You Only Look Once version 3 (YOLOv3) algorithm with and without transfer learning is applied on the Karlsruhe Institute of Technology and Toyota Technological Institute (KITTI) dataset for cars, cyclists, and pedestrians detection. This application shows that objects could be detected in real time and the transfer learning boosts the detection performance. The Convolutional Recurrent Neural Network (CRNN) algorithm with a pre-trained model is applied on multiple License Plate (LP) datasets for real-time LP recognition. The optimized model is then used to recognize Ontario LPs and achieves high accuracy. The Efficient Residual Factorized ConvNet (ERFNet) algorithm with transfer learning and a cubic spline model are modified and implemented on the TuSimple dataset for lane segmentation and interpolation. The detection performance and speed are comparable with other state-of-the-art algorithms. / Thesis / Master of Applied Science (MASc)