Aiming at the problem of low vehicle detection rate in street scene, a forward vehicle target detection algorithm based on improved Faster R-CNN is proposed. Firstly, the deep linear convolutional neural network is used to fully extract the target features; then, for the difficult to detect vehicle targets, the difficulty of mining is introduced in the regional recommendation network to make the training more adequate, and the clustering algorithm is used to determine the length and width ratio of the recommendation frame; In the small target detection problem, the RoI normalization algorithm of bilinear interpolation is introduced into the pooling layer of the target area. Finally, an adaptive optimization algorithm is selected to optimize the parameters. The experiments were performed on the KITTI dataset for training and testing. The results show that the average accuracy of the improved model is 77.1%, 7.39% higher than the original Faster R-CNN, the total training time is reduced by 2.95h, and the total test time is reduced by 9.724s. In this case, the training and detection time is reduced, and the requirements for improving the detection performance of the vehicle ahead can be met.