Facing the technical bottlenecks in ensuring the safety and reliability of environmental perception in autonomous driving, this paper systematically reviews the technological roadmap and recent developments of LiDAR-camera fusion-based object detection. By analyzing the limitations of single-sensor perception under complex road conditions, the study demonstrates the inevitability of multimodal sensor fusion. The findings indicate that sensor fusion can significantly improve object detection performance in scenarios involving occlusion, long distances, and complex environments. Experiments on public datasets such as nuScenes and KITTI further verify that sensor fusion effectively enhances perception robustness and detection accuracy in diverse scenes. However, current technologies still face challenges such as insufficient real-time performance, high algorithmic complexity, and cost constraints. Future research should focus on improving the precision and efficiency of fusion algorithms and promoting the standardization of multimodal collaborative perception.

To avoid the missing detection phenomenon caused by loss of local features due to the occlusion and truncation under complex traffic scenes, a LiDAR-based 3D object detection algorithm based on an improved PointPillars framework is proposed. First, the point cloud feature encoding network in PointPillars is enhanced to better preserve important structural information and the relationships within the point cloud data. Then, multi-scale point cloud features are extracted and aggregated through stacked sparse convolution modules and submanifold convolution modules. On the bird's-eye-view feature map that fuses these multi-scale features, 3D bounding boxes of objects are classified and regressed, improving the network's operational efficiency. Finally, experimental tests are conducted on the KITTI dataset. The results show that compared to the traditional PointPillars algorithm, the proposed 3D object detection method improves the Average Precision (AP) of 3D detection for challenging vehicle samples by 6.02 percentage points, and increases the AP for simple pedestrian and cyclist samples in the bird's-eye view by 3.35 and 3.18 percentage points respectively, effectively enhancing 3D object detection accuracy.

To enhance the accuracy of environmental perception and object detection for intelligent vehicles, this paper proposes a novel feature fusion-based 3D object detection method. First, the Transformer model is improved by leveraging a multi-head self-attention mechanism to fully capture local spatial feature information while preserving diverse feature weights, enabling the extraction of point cloud and image features. An Unscented Kalman Filter (UKF) is then utilized to design a multimodal fusion system that combines image and point cloud data, ultimately achieving 3D object detection. The proposed model is trained and tested on both public and private datasets and is compared against several state-of-the-art algorithms. Experimental results demonstrate that the proposed feature fusion-based 3D object detection model achieves an improvement in mean Average Precision (mAP) of 0.34 percentage points and 3.03 percentage points compared to widely used algorithms such as F-ConvNet and F-PointNet, respectively. For vehicle and cyclist detection tasks, the mAP increases by 2.52 percentage points and 9.32 percentage points, respectively. Moreover, the inference results on real-world vehicle data align closely with the training evaluation outcomes. These findings indicate that the proposed algorithm improves the accuracy of object detection in perception systems to a certain extent and provides a valuable reference for the design of object detection systems in autonomous driving.

To address the challenges faced by conventional lidar systems in accurately filtering out water spray and dust, highly reflective expansion crosstalk noise, and the suboptimal removal of large-scale outliers, a laser point cloud denoising algorithm based on ground fitting and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) is proposed. Initial noise location information is obtained through ground point cloud segmentation, an enhanced DBSCAN algorithm, and a three-dimensional bounding box estimation method. Noise is then marked and filtered out based on its geometric and physical characteristics. The experimental results demonstrate that in dust and water spray scenarios, the noise-reduced Chamfer Distance (CD) decreases from 0.037 4 m to 0.019 8 m, while the Point-to-Mesh (P2M) distance reduces from 0.158 m to 0.088 m. The overall geometric deviation of the point cloud is significantly diminished, with improved conformity to the true surface. For processing complex point cloud data, the optimised DBSCAN clustering achieves a computational time of merely 6 ms compared to the conventional DBSCAN algorithm, representing a 77.8% improvement in computational efficiency. This effectively resolves the suboptimal outlier filtering performance of traditional laser point cloud denoising algorithms at large scales.

By using the calculation method of one-dimensional-three-dimensional joint simulation, the influences of AGS layout form on the intake air volume of the front cabin, the heat dissipation of the cooling module and the energy consumption of the air conditioning compressor were explored. The research finds that when the distance between AGS blades and the grille is 57~97 mm, the effects of direct and non-direct arrangements on heat dissipation performance and energy consumption tend to be consistent. The attenuation of intake air volume is approximately 2.4%, the reduction of heat dissipation is less than 1%, and there is almost no impact on the energy consumption of the compressor. When the distance is less than 57 mm (close to the grille), the direct arrangement is superior to the staggered arrangement. This is reflected in the fact that the intake air volume increases by 3.16% compared to the staggered arrangement, the heat dissipation increases by approximately 2%, and the energy consumption of the compressor decreases by about 2%. When the distance is greater than 97 mm (close to the cooling module), the adoption of a staggered arrangement is more conducive to reducing the air intake attenuation caused by AGS blades.

A study was conducted on the collision safety of passengers with “zero gravity” seats at different deployment angles. Firstly, a frontal collision model of the vehicle is established, and the seat angles were adjusted to standard posture, zero gravity sitting posture, and the backrest angle was adjusted to three different postures of 120°, 150°, and 180° based on the standard posture. Then, two collision speed curves of 40 km/h and 56 km/h were applied to the model, respectively. Simulation comparison shows that the injury values of the occupant's head, neck, and chest are the smallest in the standard posture. The injury values of the occupant's head and neck are the largest in the zero-gravity posture. The risk of chest injury increases with the increase of posture angle. The chest undergoes compression deformation, and the lungs are most susceptible to contusion, followed by the liver and spleen. The increase in the inclination angle of the passengers increases the risk of severe diving and “secondary collision” of the lower limbs.

Based on the vehicle-vehicle oblique collision accident scenario caused by cooperative obstacle avoidance, this paper analyzes the displacement response and occupant injury. 2 variables are selected, namely the barrier residual vehicle speed and the collision overlap rate: 4 types of barrier vehicle speeds and 3 types of collision overlap rates are set respectively. The displacement phenomenon of occupants caused by cooperative obstacle avoidance and occupants injuries during the collision process are analyzed; the active pre-tensioning seat belt is matched and its restraint effect on the displaced occupants is analyzed. The results show that cooperative obstacle avoidance can cause obvious lateral displacement of the occupants, reducing the protective effect of the basic restraint system, especially for right-leaning seated occupants, the seat belt has completely detached from the occupants' shoulders. From the analysis results of the test matrix, the occupant will suffer the highest comprehensive damage when the collision overlap rate is about 30%. The restraint system equipped with active seat belt has good restraint effect on the displacement movement of the occupants during the cooperative obstacle avoidance process. The lateral displacement of the occupants is significantly reduced, and the comprehensive injury risk of the occupants decreases, whereas there is still a high risk of injury to the occupants' heads in the collision condition with a 50% overlap rate.

Full Width Deformable Barrier (FWDB) finite element model is established according to the standard of EEVC-WG15, calibrated using the Transport Research Laboratory (TRL) trolley experiment method, and the Moving Progressive Deformable Barrier (MPDB) finite element model that has been calibrated and verified is used to build the MPDB and FWDB crash models of cars and SUVs respectively for the frontal collision compatibility study of vehicles. Compatibility evaluation indexes of MPDB and FWDB models are calculated, the analysis results show that the current MPDB conditions have some limitations on the evaluation of structural compatibility of vehicles, and cannot evaluate the role of the primary energy absorbing structure in the vertical and horizontal directions, and need to be combined with the compatibility evaluation indexes of FWDB conditions, VSI and HSI, to conduct an analysis in order to evaluate the frontal collision compatibility of the vehicle in a more comprehensive way.

In order to explore the influence of ball hinge strength on small offset collision in chassis structure and improve the accuracy of simulation model in practical engineering application, test and simulation are adopted to obtain the corresponding material mechanical parameters through the material real test of key components of chassis and the ball head failure test, and the ball hinge strength at key positions is designed. A local trolley is built to calibrate the chassis parts to improve the simulation accuracy of the analysis model. The results show that the strength of the ball hinge has a great influence on the small offset collision condition, and the model analysis results after calibration are in good agreement with the test.

To investigate the impact of full-touch screen infotainment system response speed on car-following safety, this paper conducts human-machine interaction experiments using real car-following videos in an in-vehicle cockpit simulator. Data are collected from 25 participants under 2 vehicle speeds and 3 response speed conditions during both static and dynamic car-following scenarios. The visual distraction data are analyzed to reveal correlations between system response speed and car-following safety risks. Results demonstrate that as response time increases from 0 to 800 ms, the Total Time Spent on Screen (TTS), Glance-to-Screen Count (GSC), and Single Operation Time (SOT) exhibit nonlinear gradually growth with decelerating rates. Specifically, at 80 km/h, when response time increases from 0 to 400 ms, the TTS, GSC, and SOT increase by 17.83%, 19.74%, and 28.72% respectively, while the smaller increments of 10.97%, 12.89%, and 18.06%. When preceding vehicles braked at a 3 s time headway, the minimum following distance decreases by 14.5 m (80 km/h) and 12.78 m (100 km/h) as response time extends from 0 to 800 ms. Simulation results across 3 time headway levels confirm that slower system response significantly increases both collision probability and severity. These findings provide theoretical foundations for designing safer full-touch HMI systems, highlighting the critical role of optimizing response speed in mitigating driving risks.