|
|
Steps Towards Improvements of Computer Vision Methods for Traffic Analysis
Špaňhel, Jakub ; Sablatnig, Robert (oponent) ; Šikudová, Elena (oponent) ; Herout, Adam (vedoucí práce)
The rapid urbanization and increasing number of vehicles on the roads have stretched traditional traffic management systems to their limits. Intelligent Transportation Systems (ITS) offer a solution, utilizing advanced technologies to enhance traffic flow and safety. The robustness of computer vision methods within ITS, essential for traffic analysis, remains a crucial area for improvement. This thesis substantially contributes to this field, specifically focusing on Vehicle Fine-Grained Recognition, Vehicle Re-Identification, License Plate Recognition, and Monocular Vehicle Speed Measurement. Several new datasets, highly appreciated by the research community, were introduced, enhancing the evaluation and exploration within each domain mentioned earlier. The main contributions can be summarized as follows: Novel method for aggregation of visual features for vehicle re-identification & dataset. Innovative approach to license plate recognition using alignment of the license plate and holistic recognition & three published datasets. Novel augmentation techniques for vehicle fine-grained recognition & extension of previously published dataset. The biggest dataset for vehicle speed measurement & baseline evaluation with state-of-the-art methods. The key findings of this work demonstrate a significant enhancement in the accuracy, efficiency, and robustness of computer vision methods applied to traffic analysis. This research's contributions have been recognized at top conferences and journals in ITS, setting new standards for future work. By advancing the current state of ITS and contributing valuable resources for ongoing research, this thesis represents a step towards more sustainable and efficient intelligent transportation systems.
|
|
|
Klasifikace typů vozidel metodou dynamického borcení času
Halachkin, Aliaksei ; Honec, Peter (oponent) ; Honzík, Petr (vedoucí práce)
Tato práce se věnuje metodě borcení času. Během práce byla napsaná C/Python knihovna, která je použita na klasifikaci typů vozidel podle profilů. Testování se provádělo na reálných datech z laserového skeneru. Algoritmus byl porovnán s korelací a Euklidovskou vzdáleností. Nakonec byl vytvořen laboratorní přípravek, který demonstruje rozpoznávání vozidel metodou borcení času.
|
|
|
Fine-Grained Recognition and Re-Identification of Vehicles Using Advanced Feature Extraction
Doseděl, Ondřej ; Hradiš, Michal (oponent) ; Špaňhel, Jakub (vedoucí práce)
The aim of this theses was to analyze and improve methods used for fine-grained vehicle recognition and vehicle re-identification. The proposed method can be used both for recognition and re-identification. It was based on 3D bounding boxes, which were used to detect the vehicle on the image and then the vehicle was normalized by unpacking into 2D. Improvement of this method was done by determining direction of the vehicle and distinguishing between front and rear while unpacking the vehicle. This proposed method improved the existing method based on 3D bounding boxes for recognition, reducing error up to 13 % in single sample accuracy and up to 17 % track accuracy. However, no improvement was gained for vehicle re-identification using LFTD aggregation.
|
|
|
Fine-Grained Recognition and Re-Identification of Vehicles Using Advanced Feature Extraction
Doseděl, Ondřej ; Hradiš, Michal (oponent) ; Špaňhel, Jakub (vedoucí práce)
The aim of this theses was to analyze and improve methods used for fine-grained vehicle recognition and vehicle re-identification. The proposed method can be used both for recognition and re-identification. It was based on 3D bounding boxes, which were used to detect the vehicle on the image and then the vehicle was normalized by unpacking into 2D. Improvement of this method was done by determining direction of the vehicle and distinguishing between front and rear while unpacking the vehicle. This proposed method improved the existing method based on 3D bounding boxes for recognition, reducing error up to 13 % in single sample accuracy and up to 17 % track accuracy. However, no improvement was gained for vehicle re-identification using LFTD aggregation.
|
|
|
Klasifikace typů vozidel metodou dynamického borcení času
Halachkin, Aliaksei ; Honec, Peter (oponent) ; Honzík, Petr (vedoucí práce)
Tato práce se věnuje metodě borcení času. Během práce byla napsaná C/Python knihovna, která je použita na klasifikaci typů vozidel podle profilů. Testování se provádělo na reálných datech z laserového skeneru. Algoritmus byl porovnán s korelací a Euklidovskou vzdáleností. Nakonec byl vytvořen laboratorní přípravek, který demonstruje rozpoznávání vozidel metodou borcení času.
|
host ::
RSS.