In order to effectively enhance the accuracy of obstacle detection in unmanned driving on roads, this paper proposes an improved Faster-RCNN object detection model. Diverging from conventional Faster-RCNN models that replace the feature extraction network with ResNet50 instead of VGG16 and deepen the convolutional layers, allowing for a more comprehensive utilization of feature information. The proposed model is trained and tested in comparison with the EfficientNet network on the same dataset, VOC2007. Experimental results indicate that the proposed model exhibits higher precision in detecting obstacles on roads, showcasing broad applicability and achieving effective target recognition.