Perspective Aware Road Obstacle Detection

Abstract

While road obstacle detection techniques have become increasingly effective, they typically ignore the fact that, in practice, the apparent size of the obstacles decreases as their distance to the vehicle increases. In this paper, we account for this by computing a scale map encoding the apparent size of a hypothetical object at every image location. We then leverage this perspective map to (i) generate training data by injecting onto the road synthetic objects whose size corresponds to the perspective foreshortening; and (ii) incorporate perspective information in the decoding part of the detection network to guide the obstacle detector. Our results on standard benchmarks show that, together, these two strategies significantly boost the obstacle detection performance, allowing our approach to consistently outperform state-of-the-art methods in terms of instance-level obstacle detection.

A Novel Method for Road Anomaly Objects Detection in the Traffic Environment With Multi-Mechanism Fusion

“…In order to evaluate the detection effect of the detection method MMF (Multiple Mechanism Fusion) in this paper on road anomaly objects, the method in this paper is compared with the following three methods: 1) DeeplabV3+ [6]: it is a method that utilizes the encoding-decoding structure and the dilated Convolution to enhance the feature extraction effect on objects of different sizes; 2) Resynthesis [27]: This is a method of detecting anomalies by comparing the differences between the input image, the semantic segmentation image, and the generated image through the dissimilarity network; 3) Softmax Entropy [17]: It is a baseline method that measures uncertainty from the predicted softmax distribution and classifies samples as out-of-distribution samples through simple statistics. 4) Road Inpainting [30]: In this approach, the road drivable areas are first patched entirely, and then a dissimilarity network is utilized to identify the discrepancies between the original image and the patched image, thereby indicating the presence of anomaly objects that have been erased by the patches. 5) JSRNet [26]: This method employs a reconstruction module to identify and reconstruct road surfaces.…”

“…This integration incorporates information from known classes and generates the final per-pixel anomaly scores for anomaly identification. With the rise of generative adversarial networks, new methods [27]- [30] have utilized these networks to create new input images based on the semantic feature maps generated by the semantic segmentation model, making better use of the feature differences between input images and synthesized images. The advantage of dissimilarity network methods lies in not excessively relying on the segmentation quality of the segmentation network.…”