- 期刊
Static Vehicle Detection Based on Improved Faster RCNN
摘要
Faster RCNN convolutional neural network to study the problems existing in vehicle detection: 1. Different scales of vehicle, especially when the remote vehicle detection rates are low. 2. The shape and size of the vehicle would change due to motion or camera, which makes it impossible to detect the vehicle. 3. When the vehicle overlaps with other things or the distance is too close, the frame regression algorithm leads to missing detection. In this paper, the convolution neural network model is improved by introducing the multi-connection feature pyramid model, variable convolution and softening non-maximum linear suppression algorithm, and then the online difficult case mining strategy and multi-scale training strategy is introduced to improve the training strategy. The experimental results show that the average accuracy (mAP) of the improved faster RCNN on COCO2019 data set reaches 54.8%, which is nearly 11.4% higher than that of the unimproved model, and it is superior to other mainstream detection networks in detection accuracy.