FAMN: Feature Aggregation Multipath Network for Small Traffic Sign Detection

Abstract

Traffic sign detection has achieved promising results in recent years. Nevertheless, there are still two problems remain to be overcome. One problem is the detection of small traffic signs, which usually occupy less than 2% of the image area. The other problem is fine-grained classification, with difficulties arising from similar appearances between traffic signs. For example, different speed-limit traffic signs have differences solely from the speed numbers. In this paper, we propose a Feature Aggregation MultiPath Network (FAMN) to tackle the problems simultaneously. First, we propose a Feature Aggregation (FA) structure to aggregate regional features from different feature maps by using element-wise Max, then convolution layers are used to extract rich semantic features. Accordingly, objects of different scales can choose the best features to improve performance of small object detection. Second, we propose a Multipath Network (MN) structure to obtain fine-grained features. The MN structure consists of three paths, extracting instance-level, part-level, and context-level features, respectively. The three types of features are then concatenated to form fine-grained features of the proposals. Experimental results demonstrate the effectiveness of our proposed FAMN. Specifically, FAMN is able to obtain an average F1-measure of 93.1% in TT100K dataset, 2.9% higher than the state-of-the-art.

Traffic Sign Detection and Recognition Using Novel Center-Point Estimation and Local Features

“…References [15], [34]- [36] explore attention mechanisms to improve performance of small traffic signs. References [16], [17] use multi-scale features which are attained from feature maps of different levels to achieve scale invariant detection. References [16], [17], [37] utilize the context information surrounding objects to increase classification accuracy.…”

“…References [16], [17] use multi-scale features which are attained from feature maps of different levels to achieve scale invariant detection. References [16], [17], [37] utilize the context information surrounding objects to increase classification accuracy. References [38], [39] achieve better detection of small objects by using GAN to generate super-resolved representations for them.…”

“…References [38], [39] achieve better detection of small objects by using GAN to generate super-resolved representations for them. Reference [17] extracts local features to achieve fine-grained classification. Reference [18] makes an effort to decrease the delay of inference.…”

“…F1-measure is also used as an additional metric for more intuitive comparisons which considers both precision and recall. We compare our method with three representative generic object detectors Faster-RCNN [8], SSD [10] and FPN [11] and four state-of-the-art traffic sign detectors Zhu et al [14], MR-CNN [16], pGAN [38] and FAMN [17]. Tab.1 shows the detection performance of our and the other seven methods in the three size groups.…”

“…MR-CNN [16] uses the multi-scale features and the surrounding context information of the candidate objects to detect traffic signs, but it ignores the important local features. FAMN [17] proposes a feature aggregation multi-path network to build fine-grained features. It works well, but it is slowed down by additional operations.…”