Metric learning aims to construct an embedding where two extracted features corresponding to the same identity are likely to be closer than features from different identities. This paper presents a method for learning such a feature space where the cosine similarity is effectively optimized through a simple re-parametrization of the conventional softmax classification regime. At test time, the final classification layer can be stripped from the network to facilitate nearest neighbor queries on unseen individuals using the cosine similarity metric. This approach presents a simple alternative to direct metric learning objectives such as siamese networks that have required sophisticated pair or triplet sampling strategies in the past. The method is evaluated on two large-scale pedestrian re-identification datasets where competitive results are achieved overall. In particular, we achieve better generalization on the test set compared to a network trained with triplet loss.

In this paper, we propose an integrated system to detect and track a single operator that can switch off and on when it leaves and (re-)enters the scene. Our method is based on a set-valued Bayes-optimal state estimator that integrates RGB-D detections and image-based classification to improve tracking results in severe clutter and under long-term occlusion. The classifier is trained in two stages: First, we train a deep convolutional neural network to obtain a feature representation for person re-identification. Then, we bootstrap a classifier that discriminates the operator from remaining people on the output of the state-estimator. We evaluate the approach on a publicly available multi-target tracking dataset as well as custom datasets that are specific to our problem formulation. Experimental results suggest reliable tracking accuracy in crowded scenes and robust re-detection after long-term occlusion.

Simple Online and Realtime Tracking (SORT) is a pragmatic approach to multiple object tracking with a focus on simple, effective algorithms. In this paper, we integrate appearance information to improve the performance of SORT. Due to this extension we are able to track objects through longer periods of occlusions, effectively reducing the number of identity switches. In spirit of the original framework we place much of the computational complexity into an offline pre-training stage where we learn a deep association metric on a largescale person re-identification dataset. During online application, we establish measurement-to-track associations using nearest neighbor queries in visual appearance space. Experimental evaluation shows that our extensions reduce the number of identity switches by 45%, achieving overall competitive performance at high frame rates.

In this paper we describe an approach for detection and pose estimation of colored objects with only few or no textural features. The approach consists of two separate stages. First, we perform vision-based object detection and hypothesis filtering. Then, we estimate and validate the object's pose in 3-D laser scans. For object detection we integrate image segmentation results from multiple viewpoints in a set-theoretical filter that provides a probabilistically sound estimate of the number of objects and their respective locations. For validation and pose estimation we search for the best pose by sampling from a geometric measurement model. The system has been validated during autonomous exploration missions in unstructured and space-like environments.

The Probability Hypothesis Density (PHD) filter is an efficient formulation of multi-target state estimation that circumvents the combinatorial explosion of the multi-target posterior by operating on single-target space without maintaining target identities. In this paper, we propose a multi-target tracker based on the PHD filter that provides instantaneous state estimation and delayed decision on data association. For this purpose, we reformulate the PHD recursion in terms of single-target track hypotheses and solve a min-cost flow network for trajectory estimation where measurement likelihoods and transition probabilities are based on multi-target state estimates. In this manner, the presented approach combines global data association with efficient multi-target filtering. We evaluate the approach on a publicly available pedestrian tracking dataset to present state estimation and data association capabilities.

Detection and tracking of pedestrians is an essential task for autonomous outdoor robots. Modern 3D laser range finders provide a rich and detailed 360 degree picture of the environment. Unstructured environments pose a difficult scenario where a variety of objects with similar shape to a human like shrubs or small trees occur. Especially in combination with partial occlusions, sensor noise, and conclusions from traversing rough terrain.

The detection and tracking of moving vehicles is a necessity for collision-free navigation. In natural unstructured environments, motion-based detection is challenging due to low signal to noise ratio. This paper describes our approach for a 14 km/h fast autonomous outdoor robot that is equipped with a Velodyne HDL-64E S2 for environment perception. We extend existing work that has proven reliable in urban environments. To overcome the unavailability of road network information for background separation, we introduce a foreground model that incorporates geometric as well as temporal cues. Local shape estimates successfully guide vehicle localization. Extensive evaluation shows that the system works reliably and efficiently in various outdoor scenarios without any prior knowledge about the road network. Experiments with our own sensor as well as on publicly available data from the DARPA Urban Challenge revealed more than 96% correctly identified vehicles.