Research On Target Detection Of Driverless Vehicles In Front Of Road Based On Mask R-CNN

The main target detection technology for driverless cars on the road is the perception of various complex environments in the outside world,and then make decisions about it,so that the driverless cars can operate autonomously and respond to driving.However,in the process of detecting objects outside the vehicle,because of complex external factors,the accuracy of target detection will be greatly affected,and many of the more blurred,tiny,and skewed targets in the sample are difficult to segment,and there are blocked vehicles,Road signs,parking lines and various obstacles.This article uses the target detection Mask R-CNN algorithm and uses the road ahead cityscapes dataset for training and testing,predicts the driving area ahead and detects the target in front of the road,and in the target detection for fuzzy,tiny and skewed targets.Then the key point detection method is used to improve the accuracy of Mask R-CNN.In this article,we mainly completed the following work:1.Compare Mask R-CNN with commonly used neural network algorithms by testing on the Cityscapes data set,reflecting the advantages of Mask R-CNN.2.Apply Mask R-CNN to the target detection in front of the road of driverless cars.In the prediction of the driving area in front of the road,compared with the traditional image segmentation method,it is found that Mask R-CNN predicts the driving area more Precise.3.Add key point detection in Mask R-CNN,select multiple key points for the target,make a mask for the key point position of each category,and propose a method to predict a small number of key points with multiple key points to improve Mask R-The performance of the CNN algorithm.