Stream Query Denoising for Vectorized HD Map Construction

Shuo Wang,Fan Jia,Yingfei Liu,Yucheng Zhao,Zehui Chen,Tiancai Wang,Chi Zhang,Xiangyu Zhang,Feng Zhao

2024-01-18

Abstract:To enhance perception performance in complex and extensive scenarios within the realm of autonomous driving, there has been a noteworthy focus on temporal modeling, with a particular emphasis on streaming methods. The prevailing trend in streaming models involves the utilization of stream queries for the propagation of temporal information. Despite the prevalence of this approach, the direct application of the streaming paradigm to the construction of vectorized high-definition maps (HD-maps) fails to fully harness the inherent potential of temporal information. This paper introduces the Stream Query Denoising (SQD) strategy as a novel approach for temporal modeling in high-definition map (HD-map) construction. SQD is designed to facilitate the learning of temporal consistency among map elements within the streaming model. The methodology involves denoising the queries that have been perturbed by the addition of noise to the ground-truth information from the preceding frame. This denoising process aims to reconstruct the ground-truth information for the current frame, thereby simulating the prediction process inherent in stream queries. The SQD strategy can be applied to those streaming methods (e.g., StreamMapNet) to enhance the temporal modeling. The proposed SQD-MapNet is the StreamMapNet equipped with SQD. Extensive experiments on nuScenes and Argoverse2 show that our method is remarkably superior to other existing methods across all settings of close range and long range. The code will be available soon.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

This paper attempts to solve the problem of insufficient utilization of time information in the construction process of high - precision maps (HD - maps) in the field of autonomous driving. Specifically, although streaming methods have made certain progress in time modeling, directly applying these methods to the construction of vectorized high - precision maps fails to fully exploit the potential of time information. This paper proposes a new strategy named Stream Query Denoising (SQD), aiming to promote the learning of temporal consistency among map elements in the streaming model by denoising the queries perturbed by noise. The SQD strategy generates noisy queries by adding noise to the true information of the previous frame and simulates the prediction process of stream queries through the reconstruction process to improve the effect of time modeling.

### Main contributions of the paper
1. **Explored normal query denoising for high - precision map construction**: This is the first study on this, including three different noise strategies designed for curves and their methods of obtaining noisy queries.
2. **Proposed the stream query denoising strategy**: Helps the streaming model learn the temporal consistency of map elements.
3. **Developed SQD - MapNet**: Integrated the SQD strategy into StreamMapNet. Experimental results show that it significantly outperforms existing methods in existing benchmarks, both in the original and new settings.

### Method overview
- **Stream Query Denoising (SQD)**: Generates noisy queries by adding noise to the true information of the previous frame, and then simulates the prediction process of stream queries through the reconstruction process to improve the effect of time modeling.
- **Adaptive Temporal Matching (ATM)**: Solves the problem that there is no explicit one - to - one correspondence between the temporal ground truth and the current ground truth, and establishes a matching relationship by calculating the bidirectional Chamfer distance.
- **Dynamic Query Noising**: Considers the inherent noise of each instance and dynamically adjusts the attenuation rate of the noise to more accurately simulate temporal changes.

### Experimental results
- **nuScenes dataset**: In the 30 - meter and 50 - meter perception ranges, the performance of SQD - MapNet is significantly better than existing methods, reaching 63.9 and 64.0 mAP respectively, which is more than 3.0 mAP higher than the previous state - of - the - art method StreamMapNet. Using the stronger V2 - 99 backbone network, the performance is further improved to 74.0 and 75.2 mAP.
- **Argoverse2 dataset**: In the 60×30 - meter and 100×50 - meter ranges, the performance of SQD - MapNet is also significantly better than that of StreamMapNet, with an improvement of about 2.0 mAP respectively.

### Ablation experiments
- **Influence of main modules**: By gradually adding the proposed modules (such as dynamic query denoising and adaptive temporal matching), the contribution of each module to the final performance is verified. The results show that dynamic query denoising and adaptive temporal matching bring performance improvements of 3.7 mAP and 1.0 mAP respectively.
- **Different ways of adaptive temporal matching**: By adjusting the adaptive matching scale α, it is found that a moderate matching scale can balance the accuracy and robustness of matching, thus obtaining the best performance.

In conclusion, this paper significantly improves the effect of time modeling in high - precision map construction by introducing the stream query denoising strategy, providing new ideas and technical support for map construction in the field of autonomous driving.