Multi-task collaborative recognition technology for intelligent driving vehicles driven by computer vision

Engineering notes

Key topics: autonomous driving, perception. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

Introduction To develop a multi-task collaborative intelligent driving perception system enabling high-precision integrated recognition of pedestrians, roads, and vehicles, this study proposes a corresponding recognition technology. Methods For pedestrian detection: multi-scale dynamic binning coding and cross-modal attention fusion architecture. For road segmentation: lightweight network with hybrid attention convolution. For vehicle perception: 3D recognition system based on voxel feature fusion. Multi-task collaboration is achieved via dynamic task priority scheduling and multi-source result fusion calibration. Results Experiments show: pedestrian average accuracy 96.78%, average per-frame inference latency 14.82 ± 0.67 ms (80,000-frame dataset); road segmentation IoU 92.34%; 50 m vehicle positioning error 0.23 m; multi-target comprehensive accuracy 89.76%; complex-scene false detection rate 1.23%, missed detection rate 0.89%; 8-h continuous operation performance degradation rate 0.56%. Discussion This study enhances the accuracy and stability of intelligent driving perception in complex scenes, provides reliable perception support for autonomous driving decision-making systems, and has significant practical value for the application of multi-task collaborative sensing technology in intelligent driving.

Links and sources

• PDF from original source

Comments