An Edge Computing System with AMD Xilinx FPGA AI Customer Platform for Advanced Driver Assistance System.

The convergence of edge computing systems with Field-Programmable Gate Array (FPGA) technology has shown considerable promise in enhancing real-time applications across various domains. This paper presents an innovative edge computing system design specifically tailored for pavement defect detection within the Advanced Driver-Assistance Systems (ADASs) domain. The system seamlessly integrates the AMD Xilinx AI platform into a customized circuit configuration, capitalizing on its capabilities. Utilizing cameras as input sensors to capture road scenes, the system employs a Deep Learning Processing Unit (DPU) to execute the YOLOv3 model, enabling the identification of three distinct types of pavement defects with high accuracy and efficiency. Following defect detection, the system efficiently transmits detailed information about the type and location of detected defects via the Controller Area Network (CAN) interface. This integration of FPGA-based edge computing not only enhances the speed and accuracy of defect detection, but also facilitates real-time communication between the vehicle's onboard controller and external systems. Moreover, the successful integration of the proposed system transforms ADAS into a sophisticated edge computing device, empowering the vehicle's onboard controller to make informed decisions in real time. These decisions are aimed at enhancing the overall driving experience by improving safety and performance metrics. The synergy between edge computing and FPGA technology not only advances ADAS capabilities, but also paves the way for future innovations in automotive safety and assistance systems.

Auto-CA: Automated Cobb Angle Measurement Based on Vertebrae Detection for Assessment of Spinal Curvature Deformity.

An accurate identification and localization of vertebrae in X-ray images can assist doctors in measuring Cobb angles for treating patients with adolescent idiopathic scoliosis. It is useful for clinical decision support systems for diagnosis, surgery planning, and spinal health analysis. Currently, publicly available annotated datasets on spinal vertebrae are small, making deep-learning-based detection methods that are highly data-dependent less accurate. In this article, we propose an algorithm based on convolutional neural networks that can be trained to detect vertebrae from a small set of images. This method can display critical information on a patient's spine, display vertebrae and their labels on the thoracic and lumbar, calculate the Cobb angle, and evaluate the severity of spinal deformities. The proposed achieved an average accuracy of 0.958 and 0.962 for classifying spinal deformities (i.e., C-shaped, S-shaped type 1, and S-shaped type 2) and severity of Cobb angle (i.e., normal, mild, moderate, and severe), respectively. The Cobb angle measurement had a median difference of less than 5° from the ground-truth with SMAPE of 5.27% and an error on landmark detection of 19.73. In addition, Lenke classification is used to analyze spinal deformities as types A, B, and C, which have an average accuracy of 0.924. Physicians can use the proposed system in clinical practice by providing X-ray images via the user interface.

Fast Control for Backlight Power-Saving Algorithm Using Motion Vectors from the Decoded Video Stream.

Backlight power-saving algorithms can reduce the power consumption of the display by adjusting the frame pixels with optimal clipping points under some tradeoff criteria. However, the computation for the selected clipping points can be complex. In this paper, a novel algorithm is created to reduce the computation time of the state-of-the-art backlight power-saving algorithms. If the current frame is similar to the previous frame, it is unnecessary to execute the backlight power-saving algorithm for the optimal clipping points, and the derived clipping point from the previous frame can be used for the current frame automatically. In this paper, the motion vector information was used as the measurement of the similarity between adjacent frames, where the generation of the motion vector information requires no extra complexity since it is generated to reconstruct the decoded frame pixels before the display. The experiments showed that the proposed work can reduce the running time of the state-of-the-art methods by 25.21% to 64.22%, while the performances are maintained; the differences with the state-of-the-art methods in PSNR are only 0.02~1.91 dB, and those in power are only -0.001~0.008 W.

Touchless Head-Control (THC): Head Gesture Recognition for Cursor and Orientation Control.

The touchless techniques in human-computer interaction (HCI) can effectively expand computer access capabilities for disabled people. This paper presents Touchless Head-Control (THC), an assistive system method for computer cursor control based on head pose captured with an RGB camera. Our work aimed to replace the standard cursor control using a device on the user's head. The convolutional neural networks with predicted fine-grained feature maps and binned classification were applied to estimate the head pose angles. The mouse pointer or cursor is moved to actual locations on the screen based on head movement (yaw and pitch) and the center position of the face. Head tilt to the right or left (roll) to control the mouse button. In addition, the proposed method can be used to simulate the movement of the robot or joystick using the head to control objects within three degrees of freedom (DOF). Various participants were involved in the interaction design evaluation, in which target selection accuracy, travel time, and path efficiency were measured. This technology allows people with limited motor skills to easily control a PC cursor and 3D object orientation without the use of additional equipment or sensors.

Network-based H.264/AVC whole frame loss visibility model and frame dropping methods.

We examine the visual effect of whole frame loss by different decoders. Whole frame losses are introduced in H.264/AVC compressed videos which are then decoded by two different decoders with different common concealment effects: frame copy and frame interpolation. The videos are seen by human observers who respond to each glitch they spot. We found that about 39% of whole frame losses of B frames are not observed by any of the subjects, and over 58% of the B frame losses are observed by 20% or fewer of the subjects. Using simple predictive features which can be calculated inside a network node with no access to the original video and no pixel level reconstruction of the frame, we developed models which can predict the visibility of whole B frame losses. The models are then used in a router to predict the visual impact of a frame loss and perform intelligent frame dropping to relieve network congestion. Dropping frames based on their visual scores proves superior to random dropping of B frames.

A versatile model for packet loss visibility and its application to packet prioritization.

In this paper, we propose a generalized linear model for video packet loss visibility that is applicable to different group-of-picture structures. We develop the model using three subjective experiment data sets that span various encoding standards (H.264 and MPEG-2), group-of-picture structures, and decoder error concealment choices. We consider factors not only within a packet, but also in its vicinity, to account for possible temporal and spatial masking effects. We discover that the factors of scene cuts, camera motion, and reference distance are highly significant to the packet loss visibility. We apply our visibility model to packet prioritization for a video stream; when the network gets congested at an intermediate router, the router is able to decide which packets to drop such that visual quality of the video is minimally impacted. To show the effectiveness of our visibility model and its corresponding packet prioritization method, experiments are done to compare our perceptual-quality-based packet prioritization approach with existing Drop-Tail and Hint-Track-inspired cumulative-MSE-based prioritization methods. The result shows that our prioritization method produces videos of higher perceptual quality for different network conditions and group-of-picture structures. Our model was developed using data from high encoding-rate videos, and designed for high-quality video transported over a mostly reliable network; however, the experiments show the model is applicable to different encoding rates.