Real-time object detection using improvised YOLOv4 and feature mapping technique for autonomous driving

Drivable area detection and object detection are the primary actors in any autonomous driving technology. The proposed solution in this article contemplates a model for object detection in real-time with an improvised deep learning solution implemented with a modified YOLO algorithm. The proposed algorithm is replaced with the feature pyramid in place of spatial dimension features, for reducing the computational complexity of the standard YOLOv4 algorithm. The attention span of the YOLOv4 algorithm is improved by CBAM (Convolutional Block Attention Module) structures, where the speed of detecting the objects is associated with the keyframes and feature pyramid. BDD100K data set is available open-source and considered to be the benchmark for autonomous driving technology. This dataset is used for evaluating the performance of the proposed technique and the speed of detection significantly improved 4.72 frames per second when compared against other techniques.