Multi-Loss Function for Distance-to-collision Estimation

Estimation of the drone’s distance-to-collision is the key to the indoor autonomous obstacle avoidance and navigation of monocular UAVs. At present, the distance-to-collision model mainly uses regression loss or ordinal regression for training Regression loss utilizes the continuity of distance, and ordinal regression loss utilizes the order of distance. To improve the prediction performance of the model, this paper proposes a multi-loss function trained deep learning model based on the linear combination of ordinal regression loss and regression loss. The regression loss can be obtained by adding a distance decoder after the ordinal regression estimation, without changing the original structure of the model. Finally, we test the model performance in public datasets and obtain good results.