Exploring Weak Stabilization for Motion Feature Extraction

By Dennis Park, C. Lawrence Zitnick, Deva Ramanan, Piotr Doll'ar. 2013

142 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Signal Processing

How researchers see this paper

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

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

Delving into Egocentric Actions (2015, neutral): “Several recent efforts addressed the issue in a surveillance setting by either stabilizing the input video [25] or compensating the optical flow [37, 12].”
Video Object Detection with an Aligned Spatial-Temporal Memory (2017, methodology): “Later work used hand-designed features by aggregating simple motion cues (based on optical flow, temporal differences, or tracking), and focused mostly on pedestrian detection [45,10,23,35].”
What Can Help Pedestrian Detection? (2017, background): “, optical flow [1] and motion [29]) have been proved to be beneficial to traditional pedestrian detectors [28, 23] working on videos.”
Single-Pedestrian Detection Aided by Multi-pedestrian Detection (2013, methodology): “approaches on the Caltech-Test dataset is the ACF+SDt in [18] with the motion feature in [63], which has an average Fig.”
How Far are We from Solving Pedestrian Detection? (2016, background): “It also shows that optical flow [18] and context information [16] are complementary to image features and can further boost detection accuracy.”
A Robust and Efficient Video Representation for Action Recognition (2015, unknown): “Recently, Park et al (2013) perform weak stabilization to remove both camera and object-centric motion using coarse-scale optical ﬂow for pedestrian detection and pose estimation in video.”
Deep Learning Strong Parts for Pedestrian Detection (2015, methodology): “In the second aspect, we compare the overall result of DeepParts with existing best-performing methods , including VJ [33], HOG [4], MT-DPM [38], MT-DPM+Context [38], JointDeep [23], SDN [17], ACF+SDT [28], InformedHaar [42], ACF-Caltech+ [21], SpatialPooling [27], LDCF [21], AlexNet+ImageNet [14], ”
Real-Time Pedestrian Detection with Deep Network Cascades (2015, unknown): “We tested on pedestrians Dataset Avg. miss rate FPS Roerei [5] 46.13 1 MOCO[7] 45.5 WordChannels [9] 42.3 16 JointDeep[29] 39.3 MT-DPM+Context [40] 37.64 ACF+SDt [34] (w. motion) 37.3 SDN [25] 37.8 0.67 LFOV [3] 35.8 3.6 InformedHaar [42] 34.6 0.63 Hosang et al. [22] 32.4 SpatialPooling [31] 29.0 0.”
Filtered channel features for pedestrian detection (2015, methodology): “For our ﬂow experiments we will use difference of frames from weakly stabilized videos ( SDt ) [29].”
Ten Years of Pedestrian Detection, What Have We Learned? (2014, methodology): “They consider stereo images [45], optical ﬂow (using previous frames, e.g. MultiFtr+Motion [22] and ACF+SDt [42]), tracking [46], or data from other sensors (such as lidar [47] or radar).”
Pedestrian detection with a Large-Field-Of-View deep network (2015, methodology): “Additionally, motion information can be quite helpful, e.g. Park et al [43] utilize motion stabilization to improve the baseline classiﬁer.”
Pedestrian detection aided by deep learning semantic tasks (2014, methodology): “Features: Haar (VJ), HOG (HOG, MT-DPM), ChannelFeature (ACF+Caltech, LDCF); Classifiers: latent-SVM (MT-DPM), boosting (VJ, ACF+Caltech, SpatialPooling); Deep Models: JointDeep, SDN; Motion and context: MTDPM+Context, ACF+SDT, Katamari, SpatialPooling+. Fig.7b reports the results.”
Informed Haar-Like Features Improve Pedestrian Detection (2014, background): “ACF+SDt [22] channels + motion AdaBoost Caltech / 37.”
Pedestrian Detection: The Elephant In The Room (2020, background): “Some works also tried to incorporate extra information such as motion [29] for pedestrian detection in videos.”
Temporal Residual Networks for Dynamic Scene Recognition (2017, unknown): “[29, 30, 31].”
Semantic-Aware Scene Recognition (2019, methodology): “proposed the incorporation of opticalflow-based features to model motion through a decision-forest-based detection scheme [35].”
Computer vision and deep learning techniques for pedestrian detection and tracking: A survey (2018, methodology): “In [89] , Park et al. introduced a combined approach for motion information extraction from video sequences.”
Local Decorrelation For Improved Pedestrian Detection (2014, background): “While in the past year the use of deep learning [25], motion features [27], and multi-resolution models [36] has brought down log-average miss rate (MR) to under 40% on the Caltech Pedestrian Dataset [10], LDCF reduces MR to under 25%.”