The driver gaze zone is an indicator of a driver’s attention and plays an important role in the driver’s activity monitoring. Due to the bad initialization of point-cloud transformation, gaze zone systems using RGB-D cameras and ICP (Iterative Closet Points) algorithm do not work well under long-time head motion. In this work, a solution for a continuous driver gaze zone estimation system in real-world driving situations is proposed, combining multi-zone ICP-based head pose tracking and appearance-based gaze estimation. To initiate and update the coarse transformation of ICP, a particle filter with auxiliary sampling is employed for head state tracking, which accelerates the iterative convergence of ICP. Multiple templates for different gaze zone are applied to balance the templates revision of ICP under large head movement. For the RGB information, an appearance-based gaze estimation method with two-stage neighbor selection is utilized, which treats the gaze prediction as the combination of neighbor query (in head pose and eye image feature space) and linear regression (between eye image feature space and gaze angle space). The experimental results show that the proposed method outperforms the baseline methods on gaze estimation, and can provide a stable head pose tracking for driver behavior analysis in real-world driving scenarios.

The human eye gaze plays a vital role in monitoring people’s attention, and various efforts have been made to improve in-vehicle driver gaze tracking systems. Most of them build the specific gaze estimation model by pre-annotated data training in an offline way. These systems usually tend to have poor generalization performance during the online gaze prediction, which is caused by the estimation bias between the training domain and the deployment domain, making the predicted gaze points shift from their correct location. To solve this problem, a novel driver’s eye gaze tracking method with non-linear gaze point refinement is proposed in a monitoring system using two cameras, which eliminates the estimation bias and implicitly fine-tunes the gaze points. Supported by the two-stage gaze point clustering algorithm, the non-linear gaze point refinement method can gradually extract the representative gaze points of the forward and mirror gaze zone and establish the non-linear gaze point re-mapping relationship. In addition, the Unscented Kalman filter is utilized to track the driver’s continuous status features. Experimental results show that the non-linear gaze point refinement method outperforms several previous gaze calibration and gaze mapping methods, and improves the gaze estimation accuracy even on the cross-subject evaluation. The system can be used for predicting the driver’s attention.

Head pose and eye gaze are vital clues for analysing a driver’s visual attention. Previous approaches achieve promising results from point clouds in constrained conditions. However, these approaches face challenges in the complex naturalistic driving scene. One of the challenges is that the collected point cloud data under non-uniform illumination and large head rotation is prone to partial facial occlusion. It causes bad transformation during failed template matching or incorrect feature extraction. In this paper, a novel estimation method is proposed for predicting accurate driver head pose and gaze zone using an RGB-D camera, with an effective point cloud fusion and registration strategy. In the fusion step, to reduce bad transformation, continuous multi-frame point clouds are registered and fused to generate a stable point cloud. In the registration step, to reduce reliance on template registration, multiple point clouds in the nearest neighbor gaze zone are utilized as a template point cloud. A coarse transformation computed by the normal distributions transform is used as the initial transformation, and updated with particle filter. A gaze zone estimator is trained by combining the head pose and eye image features, in which the head pose is predicted by point cloud registration, and the eye image features are extracted via multi-scale spare coding. Extensive experiments demonstrate that the proposed strategy achieves better results on head pose tracking, and also has a low error on gaze zone classification.

Underwater object detection is a prerequisite for underwater robots to realize ocean exploration and autonomous grasping. However, underwater detection tasks face some inevitable interference factors, such as poor imaging quality, strong environment randomness, and high organism concealment. These phenomena will lead to strong underwater background interference and weak underwater object perception, which greatly aggravates the difficulty of underwater object detection. In order to deal with the above problems, we propose an unsupervised clustering optimization-based efficient attention (UCOEA). Different from the channel-wise strategy, cross-channel strategy and channel grouping strategy, we design a channel clustering strategy, which achieves autonomous dynamic screening of channel information by using the K-Means algorithm. Same types of channel information with high redundancy are learned uniformly to share the same operation. Different types of channel information with high specificity are learned independently to avoid channel noise information interference. Different from the single spatial strategy and multiple spatial strategy, we design a spatial clustering strategy, which achieves autonomous dynamic stripping of spatial information by using the EM algorithm. This strategy can extract multiple required spatial information at one time from different spatial locations. We further assign learnable weight parameters to distinguish dominant information and auxiliary information, which can alleviate spatial noise information interference. Our strategies can better balance additional cost overhead and information processing quality, which is crucial for the proposed attention to achieve fast and accurate underwater information calibration. In order to achieve high-precision and real-time underwater object detection, we propose a combined system of UCOEA underwater adapter and one-stage YOLO detector, which can efficiently detect small, medium and large targets at the same time. Extensive experiments demonstrate the effectiveness of our work. More importantly, we publish an underwater detection dataset DLMU2024 with low image continuity and high data diversity, which provides reliable support for the rapid development of underwater detection research. Our dataset is available at https://github.com/shenxin-dlmu/DLMU2024.

Driving behavior analysis is vital for the advanced driving assistance system, aiming to improve driving behavior and decrease traffic accidents. Most existing driving behavior learning methods focus on either vehicle sensor information or driver's attention information, and provide a classification result on the current time data samples. The visualization of driving behavior on time series data samples could give an understanding and review of the driver's continuous actions. However, there has been little progress in combining the multi-modal vehicle and driver information on driving behavior learning and visualization. A multi-information driving behavior learning and visualization method with natural gaze prediction is proposed in this paper, which automatically integrates driver's gaze direction estimated from face camera, and various vehicle sensor data collected from on-board diagnostics (OBD) system. To accurately estimate the eye gaze under large head movement, a novel head pose-free eye gaze prediction method without calibration is proposed based on global and local scale sparse encoding, which treats the direction mapping as small gaze region classification. To understand driving behavior more intuitively, the latent features that represent different driving behaviors are extracted by FastICA from the fused time series data, and mapped into RGB color space for distinguished visualization. Experimental results demonstrate the effectiveness of the proposed method, and show that the proposed method performs better than the compared methods.