On-line Conservative Learning for Person Detection

Roth, Peter M.; Gräbner, Helmut; Skočaj, Danijel; Bischol, H.; Leonardis, Aleš

doi:10.1109/vspets.2005.1570919

Cited by 51 publications

(49 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They controls how aggressive the automatic training process is. Similar parameters also exist in other approaches of automatically training scene specific detectors [10,11,12,13,14,15,16]. Our approach has robustness to these parameters within certain range.…”

Section: Conclusion and Discussionmentioning

confidence: 61%

“…By calculating the frame difference as 0.5(|I t − I t−50 | + I t − I t+50 |), moving pixels inside a detection window are thresholded and counted. Similar to other self-training [15,13] or co-training [11,10,16,21] frameworks, the confident positive examples are found by thresholding L p (z) > L 0 . The larger the threshold is, the more conservative the strategy of selecting examples is.…”

Section: Confident Positive Examples Of Pedestriansmentioning

confidence: 97%

“…Some automatic labelers were designed under the cotraining framework [11,10,16,21]. Two detectors based on different types of features were trained iteratively.…”

Section: Related Workmentioning

confidence: 99%

“…Roth et al [16,17] introduced classifier grids to train a separate detector for each local region. Stalder et al [18] used tracking and manually-input scene geometry to assist labeling.…”

Section: Related Workmentioning

confidence: 99%

See 3 more Smart Citations

Automatic adaptation of a generic pedestrian detector to a specific traffic scene

2011

View full text Add to dashboard Cite

In recent years significant progress has been made learning generic pedestrian detectors from manually labeled large scale training sets. However, when a generic pedestrian detector is applied to a specific scene where the testing data does not match with the training data because of variations of viewpoints, resolutions, illuminations and backgrounds, its accuracy may decrease greatly. In this paper, we propose a new framework of adapting a pre-trained generic pedestrian detector to a specific traffic scene by automatically selecting both confident positive and negative examples from the target scene to re-train the detector iteratively. An important feature of the proposed framework is to utilize unsupervisedly learned models of vehicle and pedestrian paths, together with multiple other cues such as locations, sizes, appearance and motions to select new training samples. The information of scene structures increases the reliability of selected samples and is complementary to the appearance-based detector. However, it was not well explored in previous studies. In order to further improve the reliability of selected samples, outliers are removed through multiple hierarchical clustering steps. The effectiveness of different cues and clustering steps is evaluated through experiments. The proposed approach significantly improves the accuracy of the generic pedestrian detector and also outperforms the scene specific detector retrained using background subtraction. Its results are comparable with the detector trained using a large number of manually labeled frames from the target scene.

show abstract

Section: Conclusion and Discussionmentioning

confidence: 61%

Section: Confident Positive Examples Of Pedestriansmentioning

confidence: 97%

“…Some automatic labelers were designed under the cotraining framework [11,10,16,21]. Two detectors based on different types of features were trained iteratively.…”

Section: Related Workmentioning

confidence: 99%

“…Roth et al [16,17] introduced classifier grids to train a separate detector for each local region. Stalder et al [18] used tracking and manually-input scene geometry to assist labeling.…”

Section: Related Workmentioning

confidence: 99%

See 2 more Smart Citations

Automatic adaptation of a generic pedestrian detector to a specific traffic scene

2011

View full text Add to dashboard Cite

show abstract

“…Initially, the bounding box for the walking subject is needed for the first frame which can be derived using a Histogram of oriented Gradients (HoG) approach proposed by Roth at al. [14] for people detection.…”

Section: S(x Y α)mentioning

confidence: 99%

Recognizing people in non-intersecting camera views

Bouchrika

Carter

Nixon

2009

3rd International Conference on Imaging for Crime Detection and Prevention (ICDP 2009)

View full text Add to dashboard Cite

Many studies have now confirmed that it is possible to recognize people by the way they walk. As yet there has been little formal study of identity tracking using gait over different camera views. We present a new approach for people tracking and identification between different non-intersecting uncalibrated cameras based on gait analysis. An identification signature is derived from gait kinematics as well as anthropometric knowledge. Given the nature of surveillance data, we have developed a new feature extraction technique for finding human legs. The novelty of our approach is motivated by the latest research for people identification using gait. The experimental results confirm the robustness of our method to extract gait features in different scenarios. Furthermore, experimental results revealed the potential of our method to work in real surveillance systems to recognize walking people over different views with achieved cross-camera recognition rates of 95% and 90% for two different views.

show abstract

On‐road vehicle detection and tracking based on road context and the ambient lighting adaptive framework

Kim

2008

Int J Imaging Syst Tech

View full text Add to dashboard Cite

Despite recent active research on vision-based vehicle detection and tracking, its lack of flexibility hindered practical use in real environments. To be practical, adaptation to various illumination conditions is an essential ingredient. We propose a novel adaptation framework that can improve on this current lighting adaptation using a simple road context, feature arbiter, and a proper feature fusion scheme. In real driving environments, self-supervised online learning can efficiently segment the road and nonroad regions in front of the host vehicle. Classification into these regions is very important to generate regions of interest (ROIs) for potential vehicle position, that is, road context. It improves on system efficiency by reducing noise and processing time. In our global and local lighting models, the feature arbiter selects an appropriate daytime or nighttime detector for each ROI. And finally, an adaptive fusion framework method can robustly track by selecting or removing the distinctive visual attributes. This system was successfully tested on real road data obtained with various ambient lighting conditions.

show abstract

On-line Conservative Learning for Person Detection

Cited by 51 publications

References 25 publications

Automatic adaptation of a generic pedestrian detector to a specific traffic scene

Automatic adaptation of a generic pedestrian detector to a specific traffic scene

Recognizing people in non-intersecting camera views

On‐road vehicle detection and tracking based on road context and the ambient lighting adaptive framework

Contact Info

Product

Resources

About