Sensor Fusion Approach for Object Detection and Distance Estimation in Autonomous Vehicle Perception using Camera and Lidar
Main Article Content
Abstract
The real-time integration of LiDAR and camera sensor data plays a vital role in the field of autonomous driving. This fusion enables accurate depth estimation and object detection at various distances, significantly enhancing a vehicle’s perception capabilities. This study proposes an effective method to estimate distances between a self-driving vehicle and surrounding objects such as other vehicles, pedestrians, and traffic signs through LiDAR-camera data fusion. The methodology begins with applying Rigid Body Transformations (rotation and translation) to align the coordinate frames of the LiDAR and camera systems. This is followed by projecting 3D LiDAR points onto the 2D camera image plane using Homogeneous Coordinate Transformation (matrix multiplication). The fused data is then processed with the YOLOv5 deep learning model for object detection. Distance estimation involves associating the nearest bounding box coordinates of detected objects in the camera images with the corresponding LiDAR points. Depth values are extracted from the LiDAR data, and the Euclidean distance from the ego vehicle is computed. This sensor fusion approach is rigorously evaluated using both real-world scenarios and simulated environments. The analysis includes both quantitative and qualitative assessments. The results demonstrate significant improvements in environmental perception, with consistent and reliable depth information that supports safe autonomous navigation. The technique achieved an accuracy of approximately 82%, with a mean absolute error of 4.69 and RMSE of 5.863 in highway road scenarios. These results highlight the robustness and potential of sensor fusion for enhancing perception systems in autonomous vehicles.