Local descriptors are increasingly used for the task of object recognition because of their perceived robustness with respect to occlusions and to global geometrical deformations. Such a descriptor--based on a set of oriented Gaussian derivative filters-- is used in our recognition system. We report here an evaluation of several techniques for orientation estimation to achieve rotation invariance of the descriptor. We also describe feature selection based on a single training image. Virtual images are generated by rotating and rescaling the image and robust features are selected. The results confirm robust performance in cluttered scenes, in the presence of partial occlusions, and when the object is embedded in different backgrounds.

Local descriptors are increasingly used for the task of object recognition because of their perceived robustness with respect to occlusions and to global geometrical deformations. Such a descriptor--based on a set of oriented Gaussian derivative filters-- is used in our recognition system. We report here an evaluation of several techniques for orientation estimation to achieve rotation invariance of the descriptor. We also describe feature selection based on a single training image. Virtual images are generated by rotating and rescaling the image and robust features are selected. The results confirm robust performance in cluttered scenes, in the presence of partial occlusions, and when the object is embedded in different backgrounds.

For human-robot interaction with the intent to grip a human arm, it is necessary that the ideal gripping location can be identified. In this work, the gripping location is situated on the arm and thus it can be extracted using the position of the wrist and elbow joints. To achieve this human pose estimation is proposed as there exist robust methods that work both in and outside of lab environments. One such example is OpenPose which thanks to the COCO and MPII datasets has recorded impressive results in a variety of different scenarios in real-time. However, most of the images in these datasets are taken from a camera mounted at chest height on people that for the majority of the images are oriented upright. This presents the potential problem that prone humans which are the primary focus of this project can not be detected. Especially if seen from an angle that makes the human appear upside down in the camera frame. To remedy this two different approaches were tested, both aimed at creating a rotation-invariant 2D pose estimation method. The first method rotates the COCO training data in an attempt to create a model that can find humans regardless of orientation in the image. The second approach adds a RotationNet as a preprocessing step to correctly orient the images so that OpenPose can be used to estimate the 2D pose before rotating back the resulting skeletons.

Stereo vision is a very useful, yet challenging technology for a wide variety of applications. One of the greatest challenges is meeting the computational demands of stereo vision applications that require real-time performance. The FPGA (Field Programmable Gate Array) is a readily-available technology that allows many stereo vision methods to be implemented while meeting the strict real-time performance requirements of some applications. Some of the best results have been obtained using non-parametric stereo correlation methods, such as the rank and census transform. Yet relatively little work has been done to study these methods or to propose new algorithms based on the same principles for improved stereo correlation accuracy or reduced resource requirements. This dissertation describes the sparse census and sparse rank transforms, which significantly reduce the cost of implementation while maintaining and in some case improving correlation accuracy. This dissertation also proposes the generalized census and generalized rank transforms, which opens up a new class of stereo vision transforms and allows the stereo system to be even more optimized, often reducing the hardware resource requirements. The proposed stereo methods are analyzed, providing both quantitative and qualitative results for comparison to existing algorithms. These results show that the computational complexity of local stereo methods can be significantly reduced while maintaining very good correlation accuracy. A hardware architecture for the implementation of the proposed algorithms is also described and the actual resource requirements for the algorithms are presented. These results confirm that dramatic reductions in hardware resource requirements can be achieved while maintaining high stereo correlation accuracy. This work proposes the multi-bit census, which provides improved pixel discrimination as compared to the census, and leads to improved correlation accuracy with some stereo configurations. A rotation-invariant census transform is also proposed and can be used in applications where image rotation is possible.

No description available.

No description available.