Unmanned new energy vehicles need to rely on high-precision sensors to obtain information about the surrounding environment. In complex environments, to ensure that the vehicle can accurately detect and avoid obstacles on the road during driving, to avoid traffic accidents, an obstacle detection method for unmanned new energy vehicles based on deep learning is proposed. A large amount of road obstacle image data is collected, and the convolution neural network is used to build the depth learning model of obstacle detection. Obstacle features are extracted through layer by layer convolution and pooling operations. The collected image data is used to train the deep learning model. The back propagation algorithm is employed to fine-tune the model parameters, enabling the model to detect obstacles with enhanced precision. After the model training is completed, it can be deployed to the driverless new energy vehicle, and the image of road obstacles can be captured through the camera, and then input into the depth learning model after training for obstacle detection. The experimental results show that the distance between obstacles and measuring points in this method is less than 13 m, and the error is small, which can effectively improve the ability of autonomous navigation and the accuracy of obstacle detection.