Small-Scale Pedestrian Detection Based on Topological Line Localization and Temporal Feature Aggregation

Abstract

A critical issue in pedestrian detection is to detect small-scale objects that will introduce feeble contrast and motion blur in images and videos, which in our opinion should partially resort to deep-rooted annotation bias. Motivated by this, we propose a novel method integrated with somatic topological line localization (TLL) and temporal feature aggregation for detecting multi-scale pedestrians, which works particularly well with small-scale pedestrians that are relatively far from the camera. Moreover, a post-processing scheme based on Markov Random Field (MRF) is introduced to eliminate ambiguities in occlusion cases. Applying with these methodologies comprehensively, we achieve best detection performance on Caltech benchmark and improve performance of small-scale objects significantly (miss rate decreases from 74.53% to 60.79%). Beyond this, we also achieve competitive performance on CityPersons dataset and show the existence of annotation bias in KITTI dataset.

Semantic head enhanced pedestrian detection in a crowd

“…The comparisons on CityPersons test set are reported in Table 7. First, for part-based methods, we compare with OR-CNN [9], TLL [20], Att-part [8] and Part+Grid [3]. All methods except Part+Grid use R set for training.…”

“…SAF [19] builds two subnetworks for pedestrians of large or small scales. Segmentation information [20,21,2] is adopted to enable convolution layers to learn more robust and discriminative features. ALFNet [4] proposes asymptotic localization fitting to evolve the default anchor boxes of SSD by steps into improving results.…”

Hybrid channel based pedestrian detection

“…GDFL [33] includes three components: a convolutional back-bone, a scale-aware pedestrian attention module and a zoom-inzoom-out module to identify small and occluded pedestrians. TLL-TFA [32] integrates the somatic topological line localization (TLL) networks and temporal feature aggregation for detecting multi-scale pedestrians. In SDS-RCNN [36], their RPN and Binary Classification Network (BCN) do not share features, and box-based semantic segmentation is introduced to both networks as auxiliary tasks.…”

“…Multi-Scale CNN (MS-CNN) [30] and Multibranch and High-level semantic convolutional neural Network (MHN) [31] use feature maps of multiple layers to handle objects of different scales. TLL-TFA [32] integrates somatic Topological Line Localization (TLL) network and Temporal Feature Aggregation (TFA) to detect multi-scale pedestrians. Graininess-aware Deep Feature Learning method (GDFL) [33] uses scale-aware pedestrian attention masks and a zoom-inzoom-out module to identify small and occluded pedestrians.…”

Occluded Pedestrian Detection with Visible IoU and Box Sign Predictor

“…For pedestrian detection, human part feature [10,11,12,13] is introduced to handle occlusion. In [14,15,16], segmentation feature is adopted to further boost detection performance. RPN+BF [17] replaces the second stage classifier with boosted forests.…”