This paper presents a self-driving technology that relies on a built-in sensing system to detect traffic objects. The use of three-dimensional (3D) light detection and ranging (Lidar) to assist measurement and perception is feasible. As Lidar provides high-precision positioning and is not influenced by light intensity, it can be used in bad weather and effectively minimizes the number of accidents caused by self-driving vehicles. One of the intuitive clustering methods used by the modern 3D Lidar, is known as density-based spatial clustering of applications with noise (DBSCAN). DBSCAN uses a single distance threshold to classify the entire point cloud. However, due to the non-uniform scattering characteristics of 3D Lidar, the single threshold cannot meet the current clustering accuracy requirements. Moreover, the method is very time consuming. This study proposed a fast multiresolution clustering method to improve the accuracy and reduce the computations based on the DBSCAN algorithm. We used the 360 sweeping scanning pattern of 3D Lidar to search for the angular area near the point cloud and then used variable thresholds to cluster objects with different range zones. The experimental results demonstrated that the proposed method can effectively improve the accuracy and reduce the computation time required for 3D point cloud clustering.

This paper presents a novel method for generating multi-view video. The video can be viewed without wearing a pair of glasses which is more suitable for outdoor bulletin board of kiosk applications. One of the major aspects of the new method is that the depth estimation is not required. Hence, no depth warp induced holes will be introduced. Our method first extracts feature correspondence from a Left/Right (LR) image pair and recover a set of 3D point. Then, those 3D points are projected onto the multiple virtual views. Secondly, the homography matrices are calculated for pre-warping. Finally, the pixel values of the virtual views are generated by the proposed content preserve warping with smoothness constraint.

When automatic speech recognition (ASR) and speaker verification (SV) are applied in adverse acoustic environments, endpoint detection and energy normalization can be crucial to the functioning of both systems. In low signal-to-noise ratio (SNR) and nonstationary environments, conventional approaches to endpoint detection and energy normalization often fail and ASR performances usually degrade dramatically. The purpose of this paper is to address the endpoint problem. For ASR, we propose a real-time approach. It uses an optimal filter plus a three-state transition diagram for endpoint detection. The filter is designed utilizing several criteria to ensure accuracy and robustness. It has almost invariant response at various background noise levels. The detected endpoints are then applied to energy normalization sequentially. Evaluation results show that the proposed algorithm significantly reduces the string error rates in low SNR situations. The reduction rates even exceed 50% in several evaluated databases. For SV, we propose a batch-mode approach. It uses the optimal filter plus a two-mixture energy model for endpoint detection. The experiments show that the batch-mode algorithm can detect endpoints as accurately as using HMM forced alignment while the proposed one has much less computational complexity.

A spoken dialogue system capable of maintaining a humanmachine conversation over a telephone must simultaneously maintain many dialogue sessions with different callers and third-party servers running applications over the Internet. There are three main issues to be addressed. First, the caller’s identity has to be passed between each dialogue turn. Secondly, the dialogue state and connections to third party servers have to be continuously maintained. Thirdly, the system interfaces have to follow some standards, especially Internet protocols. We use session tickets and session object hashes to address these issues. Each dialogue turn is encoded in a session data object and VXML page. The session tickets contain the session ID and security-related information passed between the VoiceXML interpreter and document server. The session object hash stores the third party connection handle, which is retrieved for subsequent dialogue turns.

In this work a method for splitting continuous mixture density hidden Markov models (HMM) is presented.The approach combines a model evaluation measure based on the Maximum Mutual Information (MMI) criterion with subsequent standard Maximum Likelihood (ML) training of the HMM parameters.Experiments were performed on the SieTill corpus for telephone line recorded German continuous digit strings.The proposed splitting approach performed better than discriminative training with conventional splitting and as good as discriminative training after the new splitting approach.

An error concealment scheme for Motion Picture Experts Group (MPEG) video networking is proposed. During video transmission, video packet loss occurs in the presence of network congestion and buffer overflow. This phenomenon of packet loss transforms into single or multiple macroblock loss in the decoding process, which can severely degrade the viewing quality. The new method differs from the conventional concealment by its exploitation of temporal redundancies in a large scale. In many applications where there is significant camera motion (e.g., remote surveillance), the evolution from frame to frame can be well described by the affine motion. The motion estimation is carried out by registering images within a multiresolution pyramid. The global motion is estimated in the lowest resolution level, and is then used to update and refine the local motion. The local motion is further refined iteratively at higher resolution levels. An affine transform is used to extract translation, scaling and rotation parameters.