Detection and Tracking of Occluded People

By Siyu Tang, Mykhaylo Andriluka, Bernt Schiele. 2014

237 citations recorded by the source corpus.

We consider the problem of detection and tracking of multiple people in crowded street scenes. State-of-the-art methods perform well in scenes with relatively few people, but are severely challenged by scenes with many subjects that partially occlude each other. This limitation is due to the fact that current people detectors fail when persons are strongly occluded. We observe that typical occlusions are due to overlaps between people and propose a people detector tailored to various occlusion levels. Instead of treating partial occlusions as distractions, we leverage the fact that person/person occlusions result in very characteristic appearance patterns that can help to improve detection results. We demonstrate the performance of our occlusion-aware person detector on a new dataset of people with controlled but severe levels of occlusion and on two challenging publicly available benchmarks outperforming single person detectors in each case.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Learning Non-maximum Suppression (2017, background): “One line of work proposes to detect pairs of objects instead of each individual objects in order to handle strong occlusion [24, 29, 19].”
End-to-End People Detection in Crowded Scenes (2015, background): “To overcome the difficulties of predicting multiple objects in close proximity, several attempts have been made to jointly predict constellations of objects [5, 23, 15].”
Object detection by labeling superpixels (2015, methodology): “In [19, 48], spatial models are used to inference the occlusion.”
Subgraph decomposition for multi-target tracking (2015, background): “For example, it is common to first group single-frame hypotheses into tracklets spanning several adjacent frames, and then combine them into complete tracks [12, 19, 24, 11, 22, 23, 2].”
Detecting People Looking at Each Other in Videos (2013, background): “However, for accurately detecting persons in more extreme cases such as when a person is mostly occluded by the other, a separately trained ‘double person detector’ might be necessary, as proposed by Tang et al. (2012).”
A Convnet for Non-maximum Suppression (2015, methodology): “Co-occurrence To better handle dense crowds or common object pairs, it has been proposed to use specialized 2-object detectors (Sadeghi & Farhadi, 2011; Tang et al., 2012; Ouyang & Wang, 2013), which then require a more careful NMS strategy to merge single-object with double-object detections.”
Occlusion-aware R-CNN: Detecting Pedestrians in a Crowd (2018, background): “In [48,41], a set of occlusion patterns of pedestrians are discovered to learn a mixture of occlusion-specific detectors.”
Bi-box Regression for Pedestrian Detection and Occlusion Estimation (2018, background): “Several approaches [24, 30, 26] are specially designed to handle occlusion situations in which multiple pedestrians occlude each other.”
Toward Accurate Person-level Action Recognition in Videos of Crowed Scenes (2020, background): “The CrowdDet includes three main contributions for crowded-scenes detection: (1) an EMD loss to minimize the set distance between the two sets of proposals [26]; (3).”
Learning People Detectors for Tracking in Crowded Scenes (2013, methodology): “At 95% precision it outperforms [17] by 20.”
Multiple People Tracking by Lifted Multicut and Person Re-identification (2017, background): “Recent works on multi-person tracking focus on the tracking-by-detection approach [21, 36, 35, 29, 30].”
A Simple Baseline for Pose Tracking in Videos of Crowed Scenes (2020, background): “The CrowdDet includes three main contributions for crowded-scenes detection: (1) an EMD loss to minimize the set distance between the two sets of proposals [20]; (3).”
Single-Pedestrian Detection Aided by Multi-pedestrian Detection (2013, methodology): “DPM is used [19, 30] to learn contextual cues.”
Recent Advances in Deep Learning for Object Detection (2019, background): “Also to reduce the impact of large variance in scales, Li et al. [8] proposed Scale-aware For pedestrian detection, one of the most signiﬁcant challenges is to handle occlusion [216, 217, 218, 219, 220, 221, 222, 223, 224, 225, 214, 226].”
Robust Multi-resolution Pedestrian Detection in Traffic Scenes (2013, background): “Some papers focused on special problems in pedestrian detection, including occlusion handling [46, 43, 38, 2], speed [25, 11, 18, 4, 10], and detector transfer in new scenes [42, 27].”
Deep Learning Strong Parts for Pedestrian Detection (2015, background): “[31] proposed a pedestrian detector tailored to various occlusion levels, while Ouyang et al.”
Non-maximum Suppression for Object Detection by Passing Messages Between Windows (2014, background): “For the problem of pedestrian detection, [26] proposed to learn detection models for couples of person.”
Detection in Crowded Scenes: One Proposal, Multiple Predictions (2020, methodology): “Double-person detector [37] models person pairs in the DPM [9] framework.”
Online Multi-object Tracking Using CNN-Based Single Object Tracker with Spatial-Temporal Attention Mechanism (2017, background): “Most works [21, 49, 41, 23, 45] aim at utilizing better detectors for handling partial occlusion.”