Performance Analysis of Hyperparameter Darknet-Based Model on 3D Object Detection Autonomous Driving

Engineering notes

Input data that used in the form of velodyne obtained from KITTI (Karlsruhe Institute of Technology and Toyota Technological Institute) Benchmark.

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

Autonomous driving is an artificial intelligence development technology equipped with a sensor camera called LiDAR (Light Detection and Ranging). LiDAR is the data that can retrieve the objects from the point cloud in 3D(3 Dimensions). Since the LiDAR data is used to detect the object in the autonomous driving, the method that can compute fast and accurately is needed. In this paper, investigating the complex You Only Look Once (YOLO) version 3 Tiny and complex YOLO version 4 Tiny which are Darknet-based Tiny Model are performed to detect the object based on LiDAR data for autonomous driving application. Further, the tiny models are used in order to reduce the complexity and computation time. Moreover, the method of changing the momentum value is carried out to improve object detection performance in autonomous driving. Input data that used in the form of velodyne obtained from KITTI (Karlsruhe Institute of Technology and Toyota Technological Institute) Benchmark. Output result on this research is in the form of a model that already has a bounding box in each object with the results of its performance accuracy. The analysis carried out in this research focuses on the value of momentum 0.1, 0.5, 0.9 and 1.0. The best performing model is found in Complex-YOLOv4-Tiny with 0.1 momentum which produces mAP value of 75.3%.

Links and sources

• Official / arXiv page

Comments