A closed-loop parallel testing system implements more challenging tests to accelerate evaluation and development of autonomous vehicles.

This paper presents a systematic scheme for fusing millimeter wave (MMW) radar and a monocular vision sensor for on-road obstacle detection. As a whole, a three-level fusion strategy based on visual attention mechanism and driver’s visual consciousness is provided for MMW radar and monocular vision fusion so as to obtain better comprehensive performance. Then an experimental method for radar-vision point alignment for easy operation with no reflection intensity of radar and special tool requirements is put forward. Furthermore, a region searching approach for potential target detection is derived in order to decrease the image processing time. An adaptive thresholding algorithm based on a new understanding of shadows in the image is adopted for obstacle detection, and edge detection is used to assist in determining the boundary of obstacles. The proposed fusion approach is verified through real experimental examples of on-road vehicle/pedestrian detection. In the end, the experimental results show that the proposed method is simple and feasible.

Long-tail effect, characterized by highly frequent occurrence of normal scenarios and the scarce appearance of extreme “long-tail” scenarios, ubiquitously exists in the vision-related problems in the real-world applications. Though many computer vision methods to date have already achieved feasible performance for most of the normal scenarios, it is still challenging for existing vision systems to accurately perceive the long-tail scenarios. This deficiency largely hinders the practical application of computer vision systems, since long-tail problems may incur fatal consequences, such as traffic accidents, taking the vision systems of autonomous vehicles as an example. In this paper, we firstly propose a theoretical framework named Long-tail Regularization (LoTR), for analyzing and tackling the long-tail problems in the vision perception of autonomous driving. LoTR is able to regularize the scarcely occurred long-tail scenarios to be frequently encountered. Then we present a Parallel Vision Actualization System (PVAS), which consists of closed-loop optimization and virtual-real interaction, to search for challenging long-tail scenarios and produce large-scale long-tail driving scenarios for autonomous vehicles. In addition, we introduce how to perform PVAS in Intelligent Vehicle Future Challenge of China (IVFC), the most durable autonomous driving competition around the world. Results over the past decade demonstrate that PVAS can effectively guide the collection of long-tail data to diminish the cost in the real world, and thus promote the capability of vision systems to adapt to complex environments, alleviating the impact of long-tail effect.

We propose a new subspace-based localization of far-field (FF) and near-field (NF) narrowband signals (LOFNS) without eigendecomposition impinging on a symmetrical uniform linear array, where the oblique projection operator is utilized to isolate the NF signals from the FF ones, and the procedures of computationally burdensome eigendecomposition are not required in the estimation of the NF and FF location parameters and the computation of oblique projection operator. As a measure against the impact of finite array data, an alternating iterative scheme is presented to improve the estimation accuracy of the oblique projection operator and, hence, that of the NF location parameters, where the “saturation behavior” encountered in most of localization methods is overcome. Furthermore, the statistical analysis of the proposed LOFNS is studied, and the asymptotic mean-squared-error expressions of the estimation errors are derived for the FF and NF location parameters. Finally, the effectiveness and the theoretical analysis of the proposed LOFNS are substantiated through numerical examples, and the simulation results demonstrate that the LOFNS provides remarkable and satisfactory estimation performance for both the NF and FF signals compared with some existing localization methods even with eigendecomposition.

This paper investigates the localization of multiple near-field narrowband sources with a symmetric uniform linear array, and a new linear prediction approach based on the truncated singular value decomposition (LPATS) is proposed by taking an advantage of the anti-diagonal elements of the noiseless array covariance matrix. However, when the number of array snapshots is not sufficiently large enough, the “saturation behavior” is usually encountered in most of the existing localization methods for the near-field sources, where the estimation errors of the estimated directions-of-arrival (DOAs) and ranges cannot decrease with the signal-to-noise ratio. In this paper, an oblique projection based alternating iterative scheme is presented to improve the accuracy of the estimated location parameters. Furthermore, the statistical analysis of the proposed LPATS is studied, and the asymptotic mean-square-error expressions of the estimation errors are derived for the DOAs and ranges. The effectiveness and the theoretical analysis of the proposed LPATS are verified through numerical examples, and the simulation results show that the LPATS provides good estimation performance for both the DOAs and ranges compared to some existing methods.