Too Far to See? Not Really!—Pedestrian Detection With Scale-Aware Localization Policy

Abstract

A major bottleneck of pedestrian detection lies on the sharp performance deterioration in the presence of small-size pedestrians that are relatively far from the camera. Motivated by the observation that pedestrians of disparate spatial scales exhibit distinct visual appearances, we propose in this paper an active pedestrian detector that explicitly operates over multiple-layer neuronal representations of the input still image. More specifically, convolutional neural nets, such as ResNet and faster R-CNNs, are exploited to provide a rich and discriminative hierarchy of feature representations, as well as initial pedestrian proposals. Here each pedestrian observation of distinct size could be best characterized in terms of the ResNet feature representation at a certain layer of the hierarchy. Meanwhile, initial pedestrian proposals are attained by the faster R-CNNs techniques, i.e., region proposal network and follow-up region of interesting pooling layer employed right after the specific ResNet convolutional layer of interest, to produce joint predictions on the bounding-box proposals' locations and categories (i.e., pedestrian or not). This is engaged as an input to our active detector, where for each initial pedestrian proposal, a sequence of coordinate transformation actions is carried out to determine its proper x-y 2D location and the layer of feature representation, or eventually terminated as being background. Empirically our approach is demonstrated to produce overall lower detection errors on widely used benchmarks, and it works particularly well with far-scale pedestrians. For example, compared with 60.51% log-average miss rate of the state-of-the-art MS-CNN for far-scale pedestrians (those below 80 pixels in bounding-box height) of the Caltech benchmark, the miss rate of our approach is 41.85%, with a notable reduction of 18.66%.

Taking a Look at Small-Scale Pedestrians and Occluded Pedestrians

“…In the recent few years, CNN based methods have achieved great success on object detection and pedestrian detection [53], [4], [31], [59], [1], [2], [3]. At first, CNN [26], [45] is simply acted as the feature extractor for pedestrian detection, which is fed to the shallow classifier.…”

Pedestrian Detection Based on Two-Stream UDN

“…Brazil et al [29] implemented a cascaded-phase design with a backbone network of VGG16 for progressive region proposal and pedestrian detection other than independent detection procedures with ROI systems. Zhang et al [30] and Song et al [31] employed ResNet with 50 layers for pedestrian feature extraction and obtained state-of-the-art results. Li et al [32] utilized deep feature (feature maps from ResNet-50) pyramids for multi-resolution feature extraction and realized a competitive accuracy and real-time pedestrian detection on Geforce GTX GPU.…”

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

“…Recently, an active detection model (ADM) [24] based on multi-layer feature representations, executes sequences of coordinate transformation actions on a set of initial bounding-box proposals to deliver accurate prediction of pedestrian locations, and achieve a more balanced detection performance for different scale pedestrian instances on the Caltech benchmark. However, the aboved boundingbox based methods inevitably incorporates a large proportion of uncertain background pixels (false positive) to the human pattern, while impels the instances to be predicted as false negatives.…”

“…It has been revealed in [4,24] that large-scale pedestrian instances typically exhibit dramatically different visual characteristics and internal features from the small-scale ones. For the network, pedestrian instances of different scales should have different responses at distinct feature representation layers.…”