Ablation Study of a Person Re-Identification on a Mobile Robot Using a Depth Camera

In this paper, we describe an ablation study for a person re-identification API on a mobile robot, for a closed-world setting, using only the IR gray value image of a depth camera. Previously, we have trained the state-of-the-art neural network for person re-identification with common parameters and methods. The resulting real-time application reached as closed-world setting a rank-1-accuracy of 94.78% and a mAP of 68.16%. Now, we focused on increasing the accuracy by removing and adjusting the image processing pipeline of our dataset. By these adjustments, we have reached a rank-1-accuracy of 98.56% and a mAP of 79.05%.