Fig. 4: Experimental results of 3D object detection in autonomous driving tasks.

a The experimental setting of the object detection task on KITTI. Three types of objects were detected: cars, pedestrians, and cyclists. b Numerical results of the KITTI experiment. The blue curves are the results of the baseline method. The yellow curves are the results of our approach. The proposed method can achieve more than 100 times better performance under large resistance variances on all KITTI dataset subsets (Easy, Moderate, and Hard). The shaded areas are confidence intervals. The bar charts on the right with confidence intervals show the results of statistical tests (i.e., run the task for 20 times and compare the accuracy) at σ = 0.25. The horizontal line above the bars indicates the statistical difference in the performance of our methods compared to the baseline methods (if the difference is significant, it will be marked by the *** symbol. Otherwise, the p value will be displayed). c Visualization of 3D object detection results. The top figure is the Bird’s Eye View of the ground truth detection result. The left bottom figure is the baseline method’s result. The bottom right figure is our approach’s result. Cars and cyclists are not detected by baseline methods, while the proposed method can successfully detect these objects. The safety of analog DNNs can be largely improved with our approach.