Abstract:Person detection and tracking (PDT) has seen significant advancements with 2D camera-based systems in the autonomous vehicle field, leading to widespread adoption of these algorithms. However, growing privacy concerns have recently emerged as a major issue, prompting a shift towards LiDAR-based PDT as a viable alternative. Within this domain, "Tracking-by-Detection" (TBD) has become a prominent methodology. Despite its effectiveness, LiDAR-based PDT has not yet achieved the same level of performance as camera-based PDT. This paper examines key components of the LiDAR-based PDT framework, including detection post-processing, data association, motion modeling, and lifecycle management. Building upon these insights, we introduce SpbTrack, a robust person tracker designed for diverse environments. Our method achieves superior performance on noisy datasets and state-of-the-art results on KITTI Dataset benchmarks and custom office indoor dataset among LiDAR-based trackers.
What problem does this paper attempt to address?
The problems that this paper attempts to solve are the challenges faced by LiDAR - based Person Detection and Tracking (PDT) technology in fields such as autonomous driving, industrial safety, and crowd management. Specifically:
1. **Privacy issues**:
- Although traditional 2D - camera - based PDT systems perform excellently in terms of performance, they pose a risk of privacy leakage because these systems can capture personal identity information. In contrast, LiDAR data does not contain identifiable personal information and is therefore more suitable as a privacy - protecting alternative.
2. **Robustness in noisy environments**:
- When LiDAR sensors work in complex and noisy environments, they are easily affected by noise, resulting in a decline in detection and tracking performance. Existing LiDAR - based PDT methods have not yet reached the same performance level as camera - based methods.
3. **Optimization of the tracking module**:
- The paper deeply analyzes the key components of the LiDAR - based PDT framework, including post - detection processing, data association, motion modeling, and life - cycle management, and proposes improved algorithms to enhance overall performance.
4. **Non - linear motion and low - frame - rate problems**:
- Human motion is usually non - linear, and LiDAR sensors usually operate at 5Hz, although their original data release frequency is 10Hz. Standard Kalman filters perform poorly in this case because they assume a linear motion model and Gaussian noise distribution, which often does not hold in practical applications.
To address the above challenges, the paper proposes SpbTracker, a robust LiDAR - based person tracker designed for variable environments. SpbTracker achieves better performance through the following improvements:
- **Improved data association method**: An association metric based on object similarity and size is introduced to improve the accuracy of object - to - object association across frames.
- **Adaptive system covariance**: An adaptive system covariance method based on prediction confidence is proposed to improve the state - estimation accuracy in non - linear and under - observed situations.
- **Life - cycle management**: 3D boxes with low confidence are retained to construct more complete trajectories, avoiding discarding "ghost" detections that may represent occluded objects.
- **Comprehensive evaluation**: The generalization ability and robustness of the algorithm are verified through experiments on the KITTI dataset and a custom - made indoor office dataset.
In summary, this paper aims to improve person - tracking performance in complex and noisy environments by optimizing various components in the LiDAR - based PDT framework, thereby promoting technological progress in fields such as autonomous driving, industrial safety, and crowd management.