A Deep-Driven Multi-Sensor Fusion Framework Using CNN & LSTM for Autonomous Vehicle Optimization

Engineering notes

Key topics: autonomous driving, autonomous vehicle, lidar, sensor fusion, multi-sensor fusion, radar, 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

This paper describes the design and implementation of Autosense, a multi-sensor fusion system of deep learning for autonomous systems. Existing systems of sensing are dependent upon the use of just one type of sensor, leading to degraded accuracy and higher error rates under dynamic environments. To improve perception and reliability, the system combines multiple sensors like LiDAR, radar, GPS, and camera. It employs the use of Convolutional Neural Networks (CNN) for extracting spatial features and Long Short-Term Memory (LSTM) networks for temporal synchronization. Bayesian fusion methodology is utilized to fuse multi-sensor data and balance uncertainty along with increased decision accuracy. Autoencoder-based anomaly detection, along with Isolation Forest algorithms, are included to detect anomalies of sensor patterns. Experimental results reveal AutoSense to have increased fusion accuracy, reduced response time, and lower false detection rates relative to existing Kalman filtering methods. Improved efficiency, adaptability, and real-world applicability are demonstrated by the framework for autonomous systems.

Links and sources

• Official / arXiv page

Comments