Unified convolutional neural network for direct facial keypoints detection

Case study: Performance analysis and development of robotized screwing application with integrated vision sensing system for automotive industry

“…Modern advanced sensing systems based on "smart camera sensors" have been quickly spread into many fields of industry, such as general process control, 4 general object identification and recognition, 5 quality check (Stejskal, Bayraktar, Alghamdi), [5][6][7] reading texts and codes (e.g. car identification plate), 8,9 face recognition (Park), 10 robot control via gestures 10 and general pose control of robots using visual servoing, 3 visual navigation of single/multiple mobile robot/s (Kim, Ostertag, Mikulova, Roberti), [11][12][13][14] collision detection and perception (Lehnhardt), 15 many applications in food industry, and online process control and monitoring (Kosinar) 16 using component recognition (Stipancic, Scholz-Reiter). 17,18 Smart sensors are key elements for visual inspection of parts or visual navigation of robots for assembly tasks that has been manually done by human operators before.…”

“…Due to the nature of the processed data, VGRs must be able to learn and to "understand" the scene in real world, so they are often closely related to different approaches of artificial intelligence. Since they show better results for some specific tasks, many studies are being conducted on neural networks application (Park) 10 or combination of machine learning algorithms. 2,8,10,20 While the MV market offers various types of vision systems, such as PC-based systems, Smart Camera vision system, and hybrid Smart Camera vision system, 3 for most tasks, VGR systems are sufficient to apply twodimensional (2D) MV systems based on data collection from single camera.…”

“…Since they show better results for some specific tasks, many studies are being conducted on neural networks application (Park) 10 or combination of machine learning algorithms. 2,8,10,20 While the MV market offers various types of vision systems, such as PC-based systems, Smart Camera vision system, and hybrid Smart Camera vision system, 3 for most tasks, VGR systems are sufficient to apply twodimensional (2D) MV systems based on data collection from single camera. 15,20 On the other hand, for more advanced tasks (like, e.g.…”

“…Since they show better results for some specific tasks, many studies are being conducted on neural networks application (Park) 10 or combination of machine learning algorithms. 2,8,10,20…”

Online learning and detection of faces with low human supervision

“…A wide variety of statistical methods have been proposed for object recognition using these databases in the last years. Machine learning techniques like Support Vector Machines [8,10,35], Boosting [3,31,49,50,56], Random Forests [7,13,39], and more recently Deep Learning [19,27,40] are commonly used to build classifiers robust to large intra-class variations and other artifacts such as loss of image resolution, rotations, deformations or lighting changes.…”