Slope-Track: Multiple Object Tracking on Ski Slopes

Abstract

In this paper, we introduce Slope-Track. Slope-Track is a novel multiple object tracking (MOT) dataset designed to reflect the complexities of real ski slope environments. The dataset has over 96,000 frames collected from 10 different ski resorts under various weather and visibility conditions. Slope-Track addresses significant challenges in slope monitoring, including small object sizes, occlusions, fast and irregular motion, and low appearance consistency. It is densely annotated with bounding boxes and object identities, facilitating the evaluation of detection and tracking algorithms. We analyze the dataset’s characteristics comparing it to the existing MOT datasets. The results demonstrate that Slope-Track encapsulates a combination of challenges found in other datasets. Additionally, we benchmark a range of existing tracking algorithms and propose a new module that improves motion-based association by dealing with the specific shape of trajectories along ski slopes. Our results demonstrate that incorporating appearance features can have a mixed impact, depending on how they are used within each tracking algorithm. In contrast, motion-based methods and spatial association strategies show more reliable performance. Overall, we provide a challenging benchmark for evaluating and improving multi-object tracking systems in real-world outdoor environments.

Dataset

Examples from Slope-Track dataset showing 9 of the 10 ski slopes included in the dataset. Slope-Track includes diverse weather conditions, visibility levels and viewpoints (unique camera position and camera quality)

Slope-Track consists of:

32 videos of varying lengths; 11 videos for the training set, 4 videos for the validation set, 5 videos for the test set and an additional 12 videos that can be used as an unlabeled test set.
The dataset is 3318 seconds in length with 96,498 frames (including the unannotated videos) and 555,528 bounding boxes and 888 unique ids (annotated videos).

Results

Appearance Features

Mean inter-frame cosine distance of re-ID features

Example showing the change in features for the same object in the Slope-Track dataset including bounding box size in pixels.

Appearance analysis on MOT17, DanceTrack, SportsMOT and Slope-Track reporting on the validation set. The result shows that Slope-Track has inconsistent appearance features across frames leading to the highest inter frame object dissimilarity.

Motion Analysis

Kalman Filter-based IoU on different prediction intervals

Motion analysis of MOT17, DanceTrack, SportsMOT and Slope-Track reporting on the validation set, where the detection boxes are ground-truth boxes. The result shows that Slope-Track has fast motion, variable-speed and non-linear motion comparable to SportsMOT.

Citation

        @article{Campbell2026SlopeTrack,
          title   = {Slope-Track: Multiple Object Tracking on Ski Slopes},
          author  = {Campbell, M'Saydez and Ducottet, Christophe and Muselet, Damien and Emonet, R{\'e}mi},
          journal = {Computer Vision and Image Understanding},
          pages   = {104663},
          year    = {2026},
          issn    = {1077-3142},
          doi     = {10.1016/j.cviu.2026.104663},
          url     = {https://www.sciencedirect.com/science/article/pii/S1077314226000305}
        }

License

The dataset of Slope-Track is available for non-commercial research purposes only. Please review the license file for the terms and conditions of usage.

Acknowledgement

This research work is being carried out as part of a collaborative i-Démo Regionalized project under the French government's regionalized France 2030 program. It was financed via Bpifrance by the French government, the Auvergne-Rhône-Alpes Region and Grenoble Alpes Métropole.