The latest research result of Artificial Intelligence Research Institute of EIC, “VoxelTrack: Multi-Person 3D Human Pose Estimation and Tracking in the Wild”, was published in the top international journal in the AI field, that is IEEE Transactions on Pattern Analysis and Machine Intelligence (IEEE TPAMI) with an impact factor of 16.389. The first author is Zhang Yifu, a postgraduate student of EIC.
Zhang Yifu has published paper “FairMOT: On the Fairness of Detection and Re-Identification in Multi-Object Tracking”as the first author in International Journal of Computer Vision (IJCV), also a top international journal in the field of AI with an impact factor of 7.410. IEEE TPAMI and IJCV are the two most recognized journals in computer vision. This paper is influential in the field of multiple object tracking, which has been cited over 270 times in Google Scholar and got 3.4k star in Github. The research result has been extensively applied in the industry like Baidu Smart City and Microsoft Self-service Supermarket.
The paper published in IEEE TPAMI studies multi-person 3D human pose estimation and tracking, which is one of the most important tasks in computer vision. It can be extensively applied in various scenes such as metauniverse, virtual reality, self-service supermarket and other scenes. The paper presents VoxelTrack for multi-person 3D pose estimation and tracking from a few cameras which are separated by wide baselines. It employs a multi-branch network to jointly estimate 3D poses and re-identification (Re-ID) features for all people in the environment. In contrast to previous efforts which require to establish cross-view correspondence based on noisy 2D pose estimates, it directly estimates and tracks 3D poses from a 3D voxel-based representation constructed from multi-view images. The paper first discretizes the 3D space by regular voxels and compute a feature vector for each voxel by averaging the body joint heatmaps that are inversely projected from all views. It estimates 3D poses from the voxel representation by predicting whether each voxel contains a particular body joint. Similarly, a Re-ID feature is computed for each voxel which is used to track the estimated 3D poses over time. The main advantage of the approach is that it avoids making any hard decisions based on individual images. The approach can robustly estimate and track 3D poses even when people are severely occluded in some cameras. It outperforms the state-of-the-art methods by a large margin on three public datasets including Shelf, Campus and CMU Panoptic.
The paper published in IJCV studies multi-object tracking, which can be extensively adopted in smart city and autopilot system. The paper presents a simple approach which consists of two homogeneous branches to predict pixel-wise objectness scores and re-ID features. The achieved fairness between the tasks allows to obtain high levels of detection and tracking accuracy and outperform previous state-of-the-arts by a large margin on several public datasets.
窗体底端
