Abstract
Vehicle object detection is one of the important research directions in the field of computer vision. Aiming at solving the problems of low accuracy, slow speed, and unsatisfactory results of using traditional methods to detect the object of the vehicle in front of the driverless car on the road, this paper proposes an improved YOLOv3 vehicle target detection algorithm which we name it F-YOLOv3. First the multi-scale prediction network model is improved according to actual traffic conditions and efficiency requirements based on the original general object detection YOLOv3 algorithm. Then a scale prediction layer is added to improve the detection accuracy of large vehicles and improved k-means++ the algorithm is used to improve the effect of anchor box dimensional clustering and the detection speed. At last an experiment was conducted on a self-made dataset and compared with YOLOv3 in order to test the effectiveness of the F-YOLOv3 algorithm. The test results show that the improved F-YOLOv3 model has a precision mAP of 91.12% and a speed of 59FPS, which are better than the traditional general object detection YOLOv3 algorithm. Therefore, the algorithm has better performance and popularization prospect in vehicle object detection.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call