Autonomous driving has become the focus of research and development in recent years. However, the capability of precise positioning and obstacle avoidance for autonomous driving often rely on accurate and dense point cloud images. In particular, the environment in the metropolitan area is relatively complex and there are many viaducts and expressways in modern cities. The spatial features obtained from flat roads are no longer sufficient. If we can obtain the features of higher buildings or higher landforms, a better positioning information and security can be provided when autonomous vehicles are driving on expressways or elevated roads. In addition, the high density of vehicles in the metropolitan area and vision blind spots of vehicles also form a great safety concern for autonomous driving. In order to provide higher density point cloud information so that autonomous driving can ensure the accuracy and safety of navigation and positioning when operating in metropolitan areas, we propose in this paper a data acquisition platform base on multi-LiDAR (VLP-16) as well as a data calibration and fusion algorithm to solve the problems of vision blind spots, low vertical resolution, and sparse high-level point clouds in most of the LiDAR-based system. The proposed system adopts the industrial computer (IPC) of x86 architecture and Robotic Operating System (ROS) for overall operation. In addition, the homogeneous transformation is applied for the calibration and fusion of multi-Lidar point cloud coordinate system. Experimental results have proved that the proposed system can obtain reliable point cloud data, effectively improve the vertical resolution of the point cloud, and increase the point cloud density by more than three times, which can effectively improve the positioning reliability of autonomous driving in metropolitan areas.

Abstract This paper designs a novel method to reduce the coding complexity of 3D‐HEVC encoder by utilizing the properties of human visual perception. Two vision‐oriented edge detections are proposed: for colour texture detection, the authors adopt the Just‐Noticeable Distortion (JND); for depth map, the authors combine the Sample Adaptive Offset (SAO) and the Just Noticeable Depth Difference (JNDD) model. The authors also analyse the properties of colour texture and depth map to classify the coding tree unit (CTU) into various kinds of types, including complex‐edge CTU, moderate‐edge CTU and homogeneous CTU. Besides, fast mode decisions and early termination criteria are performed individually on each type of CTUs according to their characteristics. Especially for those CTUs with more edge information, the proposed projection‐based fast mode decision and residual‐based early termination preserve important colour texture while speeding up the coding at the same time. The proposed vision‐oriented algorithm reduces 31.981% of the overall average coding time with only 1.580% BD‐Bitrate increase. Experimental results show that the proposed algorithm can provide considerable time‐saving while still maintain the video quality, which outperforms the previous researches.

3D-HEVC (The 3D Extension of High Efficiency Video Coding) is the newest 3D video coding standard, which enriches multimedia applications with the video format of multi-view plus depth. For the depth map coding in 3D-HEVC, the advanced coding tools enhance the coding efficiency of the depth map and the quality of the synthesized view. However, the time consumption and complexity of 3D-HEVC also increase significantly. This paper utilizes the characteristics of human visual system to propose a fast algorithm based on visual perception for the acceleration of the depth intra coding of 3D-HEVC. The depth map is segmented into different regions by Otsu's auto-thresholding. The dominate edge direction is categorized for each prediction unit. We detect the perceptual edge based on just noticeable depth difference model to extract the area that may affect the visual perception. According to depth map segmentation and edge distribution, we reduce the corresponding intra angular modes and determine whether to perform depth modelling mode. We also incorporate the boundary continuity and rate-distortion cost thresholding to propose the fast coding unit decision. The experimental results show that the proposed algorithm eliminates 53.09% of the depth coding time with only 0.15% BD-BR on average. The coding performance of the proposed algorithm outperforms the previous works significantly.

We construct a robust and precise multi-orientation text detection system in scene images which can extensively locate possible characters with multi-information fusion. In our method, an adaptive multi-channel character grouping algorithm is first proposed to extract all possible character candidates robustly, and an AdaBoost classifier is then to properly identify character candidates as characters or non-characters. A single-link clustering with distance metric learning is thereafter used to adaptively group characters into text regions, and an effective hybrid filter with Convolution Neural Networks (CNN), AdaBoost and Bayesian classifiers is finally designed to precisely verify the extracted text regions. Our proposed technology is extensively evaluated on several public multi-orientation scene text datasets, e.g., MSRA-TD500 and USTB-SV1K, and is much better than state-of-the-art methods.

H.264/Advanced Video Coding (AVC) is well known for its superiority of finding an optimal tradeoff between the visual quality and the bit-rate expense. Nevertheless, the highly complex procedures of finding an optimal intra- or inter-prediction mode can degrade the run-time performance of the coding process. To speed up the run-time performance during the encoding of H.264/AVC intra-prediction mode, we apply in this paper a simple yet effective gradient evaluation approach so that the texture orientation inside the coding block can be evaluated efficiently. Moreover, we propose an adaptive selection strategy in this paper so that only a subset with a variable number of the intra-prediction modes will be sent for the rate-distortion optimization process. With the proposed gradient evaluation and adaptive selection strategy, a noticeable speedup on the run-time performance can be achieved with only a minor degradation on the visual quality and the bit-rate expense. When compared with the existing state-of-the-art fast decision algorithms, a significant improvement over prior arts on the proposed cost performance metric can be obtained, which demonstrates the superiority of the proposed approach.

We propose in this paper a novel algorithm as well as architecture for the fall accident detection and corresponding wide area rescue system based on a smart phone and the third generation (3G) networks. To realize the fall detection algorithm, the angles acquired by the electronic compass (ecompass) and the waveform sequence of the triaxial accelerometer on the smart phone are used as the system inputs. The acquired signals are then used to generate an ordered feature sequence and then examined in a sequential manner by the proposed cascade classifier for recognition purpose. Once the corresponding feature is verified by the classifier at current state, it can proceed to next state; otherwise, the system will reset to the initial state and wait for the appearance of another feature sequence. Once a fall accident event is detected, the user's position can be acquired by the global positioning system (GPS) or the assisted GPS, and sent to the rescue center via the 3G communication network so that the user can get medical help immediately. With the proposed cascaded classification architecture, the computational burden and power consumption issue on the smart phone system can be alleviated. Moreover, as we will see in the experiment that a distinguished fall accident detection accuracy up to 92% on the sensitivity and 99.75% on the specificity can be obtained when a set of 450 test actions in nine different kinds of activities are estimated by using the proposed cascaded classifier, which justifies the superiority of the proposed algorithm.

With the highly increased capability on parallel processing, computing on graphics processing units (GPUs) have been widely used in applications more than just graphics data processing. In this paper, we apply the compute unified device architecture (CUDA), a parallel computing architecture on GPUs proposed by NVIDIA, for the runtime performance enhancement in a predictively encoded lossless image compression system. For this, a least squares (LS)-adapted predictor, an effective approach for the removal of redundancy around boundaries, is applied. The adaptation process of an LS-based predictor requires multiplications of matrices for the construction of normal equations, which has been known to be the major complexity in LS adaptation process. Fortunately, matrices multiplication is most suitable to be parallel processed, which leads to the idea of speeding up the construction of normal equations with GPUs. With the proposed approach, a noticeable improvement on the runtime performance can be achieved as can be seen in the experiments.

An efficient approach for the sharpening of color images is proposed in this paper. For this, the image to be sharpened is first transformed to the Hue, Saturation, and Value (HSV) color model, and then only the channel of Value will be used for the process of sharpening while the other two channels are left unchanged. We then apply a proposed edge detector and low-pass filter to the channel of Value to pick out pixels around boundaries. After that, those pixels detected as around edges or boundaries are adjusted so that the boundary can be sharpened, and those non-edge pixels are kept unaltered. It is noted that the increment or decrement magnitude that is to be added to those edge pixels is determined in an adaptive manner based on global statistics of the image and local statistics of the pixel to be sharpened. With the proposed adaptive approach, the discontinuities can be highlighted while most of the original information contained in the image can be retained. Finally, the adjusted channel of Value and that of Hue and Saturation will be integrated to get the sharpened color image. In the proposed approach, a scaling factor can also be used for the adjustment of the additive magnitude so as to control the degree of discontinuity. Extensive experiments on natural images will be given in this paper to highlight the effectiveness and efficiency of the proposed approach.

With the fast development of network infrastructure, connecting to the Internet at any time and any place has been made easy and possible. On the other hand, as our world is suffering energy crisis on oil and natural resources shortages, how to make efficient use of limited power energy has remained a major problem to be conquered so far. Aimed to facilitate the life of human being as well as to use the limited power energy more efficiently, we propose in this paper a technology that can perform remote control and monitoring of electrical appliances on the Internet. To do this, an intelligent power socket (IPS) module that is able to control and monitoring the power of electricity is realized in this research. The IPS modules are placed in conjunction with the electrical appliances that are to be controlled from a far-end place. In addition, an embedded system-based home gateway that can be connected with the Internet is set up in which the electrical appliances are located. Moreover, the acquired power consumption information or the status of the appliances is stored in a database server in the Cloud. With the proposed structure, authorized users or system managers can log into the web server which is connected with the database, monitoring the power status and take actions on the appliances remotely. The control command from the far-end place, i.e., from the web server on the Internet, is first sent to the home gateway and then transmitted to the IPS modules through the Zigbee wireless communication protocol so that the remote control of appliances can be achieved. The proposed architecture can be easily applied to any kind of room space. Moreover, only a browser is needed for the client to communicate with the web server, no other application program is required. As the browser is now available almost on every information technology products, e.g., a notebook or a smart phone, the proposed architecture has been shown to be very convenient and useful for remote control and monitoring of electrical appliances, and hence can facilitate the life of human beings.

Based on the Grey prediction theory, we propose in this paper a two-pass algorithm for the sharpening of images. In the first pass, pixels around edges or boundaries are detected with edge detection mechanism. During the second pass, those pixels detected as around edges or boundaries are adjusted for the purpose of image sharpening, and those non-edge pixels are kept unaltered. With the proposed approach, most of the original information contained in the image can be retained. In the second pass, the magnitude, i.e., the increment or decrement, to be added to those edge pixels has to be determined. Usually, a larger additive can have a better sharpening result. However it can also lead to the saturation of intensity around edge pixels. Aimed to find the maximal additive magnitude automatically, we proposed in this paper the use of a Grey prediction model GM(1,1) so that the condition of over-sharpening in images to be sharpened can be avoided. In addition, a scaling factor can also be used for the adjustment of the additive magnitude in the proposed approach. Extensive experiments on natural images as well as medical images are also given in this paper. As we will see in the experiments, the proposed approach can have a very distinct intensity transition for pixels around edges or boundaries in the sharpened images, which demonstrates the usefulness of the proposed approach.

Edge detection is widely applied in digital image processing, especially in segmentation of images. Many of the well-known edge detectors, e.g., Sobel and Canny operator, are based on spatial filtering with a mask around the pixel under detection. In this paper, we propose a very low complexity edge detector that has two operation modes; non-causal mode and causal mode. Though simple in its form, the proposed two operation modes can pick out pixels around boundaries effectively and efficiently. For images with salt-and-pepper noise, the proposed approach offers a much better noise immunity than that of by Sobel operator. Experimental results are also given to demonstrate the superiority of the proposed approach.

Predictively encoded techniques are commonly used for lossless compression of images for its effectiveness of removing statistical redundancy between pixels. However, there can be large prediction errors for pixels around boundaries. In this paper, we introduce techniques commonly used in control systems to enhance the coding efficiency of predictive coding. Actually, the predictive coding system behaves just like a multi-input single-output system with the predictor itself can be taken as the system model. When compared with the purpose of a control system, which is to follow the system command as precisely as possible, we find the objective of both systems are the same. Moreover, an edge or a boundary among image pixels can be regarded as a step command in control systems. These observations lead to the idea of using control technologies to improve prediction result for pixels around boundaries. To realize this idea, we use an adaptive Takagi-Sugeno fuzzy neural network (TS-FNN) as the predictor. Furthermore, the widely used proportional controller in control system is implemented implicitly in the consequent part of the network so that the prediction error can be further compensated for pixels around boundaries. We find in experiments that the proposed approach can have a very good prediction result even without using any online training area for network adaptation process. This makes the proposed system more feasible under limited resources. Finally, comparisons to existing state-of-the-art lossless predictors and coders will be given to highlight the advantages of the proposed novel approach.

Many coding methods are more efficient with certain types of images than others. In particular, run-length coding is very useful for coding areas of little changes. Adaptive predictive coding achieves high coding efficiency for fast changing areas like edges. In this paper, we propose a switching coding scheme that will combine the advantages of both Run-length and Adaptive Linear Predictive coding (RALP) for lossless compression of images. For pixels in slowly varying areas, run-length coding is used; otherwise LS (least square)-adapted predictive coding is used. Instead of performing LS adaptation in a pixel-by-pixel manner, we adapt the predictor coefficients only when an edge is detected so that the computational complexity can be significantly reduced. For this, we propose an edge detector using only causal pixels. This way, the predictor can look ahead if the coding pixel is around an edge and initiate the LS adaptation in advance to prevent the occurrence of a large prediction error. With the proposed switching structure, very good prediction results can be obtained in both slowly varying areas and pixels around boundaries as we will see in the experiments.

In predictive image coding, the least squares (LS)-based adaptive predictor is noted as an efficient method to improve prediction result around edges. However pixel-by-pixel optimization of the predictor coefficients leads to a high coding complexity. To reduce computational complexity, we activate the LS optimization process only when the coding pixel is around an edge or when the prediction error is large. We propose a simple yet effective edge detector using only causal pixels. The system can look ahead to determine if the coding pixel is around an edge and initiate the LS adaptation to prevent the occurrence of a large prediction error. Our experiments show that the proposed approach can achieve a noticeable reduction in complexity with only a minor degradation in the prediction results.

We propose a switching adaptive predictor (FSWAP) with run-length encoding for lossless image coding. The proposed FSWAP system has two operation modes; run mode and regular mode. If the members in the texture context of the coding pixel have identical grey values, the run mode is used, otherwise the regular mode is used. The run mode, using run-length coding, with an arithmetic coder, is very useful for images with flat regions. The regular mode borrows the switching predictor structure in SWAP (Lih-Jen Kau et al, IEEE Trans. Fuzzy Systems) with some modifications. Experiments show that simplified context clustering is very useful in error modeling for prediction refinement. Furthermore, the execution time of FSWAP can be accelerated with minor degradation in the bit rates associated with the modifications. Comparisons of the proposed system to existing state-of-the-art predictive coders are given to demonstrate its coding efficiency.

This paper proposed a nonlinear predictor ADFK (Adaptive predictor with Dynamic Fuzzy K-means clustering error feedback) for lossless image coding based on multi-layered perceptrons. Since real images are usually nonstationary, a fixed predictor is not adequate to handle the varying statistics of input images. Using back propagation learning with causal neighbors of the coding pixel as training patterns to update network weights continuously, ADFK is made adaptive on the fly. Furthermore, prediction error is further refined in ADFK by applying error compensation different to compound context error modeling used in CALIC based on dynamic codebook design with adaptive fuzzy k-means clustering algorithm. Compensated errors are then entropy encoded using conditional arithmetic coding based on error strength estimation. The proposed compensation mechanism is proved to be very useful through experiments by further improving the bit rates in an average amount of about 0.2bpp in test images. Success in the use of proposed predictor is demonstrated through the reduction in the entropy and actual bit rate of the differential error signal as compared to that of existing linear and nonlinear predictors.

In this paper, we propose a switching adaptive predictor (SWAP) with automatic context modeling for lossless image coding. In the SWAP system, two predictors are used. For areas with edges, estimates of coding pixels are obtained using texture context matching (TCM). For all other areas, an adaptive neural predictor (ANP) is used. The SWAP encoder switches between the two predictors ANP and TCM depending on the neighborhood of the coding pixel. The switching predictor allows statistical redundancy to be removed effectively. On the other hand, it is known that prediction can be further refined using error compensation. For this, we propose the use of a modified fuzzy clustering, which leads to a modeling of errors that adapts itself to the input statistics. Experiments show that the proposed context clustering is very useful in modeling error for prediction refinement. Comparisons of the proposed system to existing state-of-the-art predictive coders will be given to demonstrate its coding efficiency.