FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W1764744555> ?p ?o ?g. }

• W1764744555 abstract "This thesis aims to advance the state of the art in pedestrian detection. Sincethere are many applications for pedestrian detection, for example automotivesafety or aiding robot-human interaction in robotics, there is a strong desirefor improvement. In this thesis, the benefits of combining multiple featuresthat gather information from different cues (for example image color, motion anddepth) are studied. Training techniques and evaluation procedures are alsoinvestigated, improving performance and the reliability of results, especiallywhen different methods are compared.While motion features were previously used, they either were conceptuallyrestricted to a setting with a fixed camera (e.g. surveillance) or were not resulting in an improvement for the full-imagedetection task. In this thesis, thenecessary modifications to the approach of Dalal et al. (which is basedon optical flow) to make it work in the full-image detection setting arepresented. In addition to this, substantial improvements using motion featuresare shown even when the camera is moving significantly, which has not beentested before. A variant of the motion feature that performs equally wellwith a significantly lower feature dimension is also introduced.Another cue that is used in the present work is color information. Usually, whenincorporating color information into computer vision algorithms, one has to dealwith the color constancy problem. In this thesis, a new feature called colorself-similarity (CSS) is introduced. It encodes long-range (between positionswithin the detector window) similarities of color distributions. By onlycomparing colors inside the detector window, the color constancy problem can becircumvented - effects of lighting and camera properties are less likely tovary significantly within the detector window than they are over the wholedataset. Additionally, it is shown that even raw color information can beuseful if the training set covers enough variability.Depth is also a useful cue. An existing stereo feature - stereo-based HOG byRohrbach et al. - is adopted and a new feature that exploits a usefulrelation between stereo disparity and the height of an object in an image isintroduced. This feature is computationally cheap and able to encode local sceneinformation, like object height and the presence of a ground plane, in acompletely data-driven way (all parameters are learned during training).It helps both by reducing false positives (eliminating those that have the wrongsize) and false negatives (those that were missed because the detector estimatedthe size wrongly).For the classifier part of the pipeline, it is shown that AdaBoost with decisionstumps is not able to handle the multi-cue, multi-view detection setting that weare examining well. A recently proposed boosting classifier, MPLBoost, turnedout to be superior, resulting in classification performance that iscomparable to support vector machines. It is also demonstrated that error ratescan be reduced by using support vector machines and boosting classifiers incombination. Another contribution of this thesis is a procedure to combinetraining datasets with different sets of cues during training, e.g. a monochrome dataset with a colored dataset, or a dataset with no motion information with a dataset fromvideo. This greatly increases the amount of available training datawhen multiple cues are used.A collection of pitfalls during evaluation is also highlighted. It isdemonstrated that the PASCAL overlap criterion encouragesoverestimating the bounding box size. Care also has to be taken when evaluatingon subsets of annotations, e.g. only on occluded pedestrians or pedestrians ofcertain sizes. When trying to determine the strengths of different approaches,naive approaches can easily lead to wrong conclusions. In this thesis, bettermethods to compare different approaches are proposed. An application of the detector in a 3D scene reasoning framework is alsopresented. Multiple detectors trained on partial (e.g. only upper body) viewsare combined. 3D reasoning is used to infer which parts of the pedestrian shouldbe visible and the framework uses this information to determine the strengths ofthe contributions of the partial detectors. This allows the detection system tofind pedestrians even when they are occluded for extended periods of time." @default.
• W1764744555 created "2016-06-24" @default.
• W1764744555 creator A5090658233 @default.
• W1764744555 date "2013-07-02" @default.
• W1764744555 modified "2023-09-26" @default.
• W1764744555 title "Multi-Cue People Detection from Video" @default.
• W1764744555 hasPublicationYear "2013" @default.
• W1764744555 type Work @default.
• W1764744555 sameAs 1764744555 @default.
• W1764744555 citedByCount "0" @default.
• W1764744555 crossrefType "dissertation" @default.
• W1764744555 hasAuthorship W1764744555A5090658233 @default.
• W1764744555 hasConcept C104114177 @default.
• W1764744555 hasConcept C115961682 @default.
• W1764744555 hasConcept C127413603 @default.
• W1764744555 hasConcept C138885662 @default.
• W1764744555 hasConcept C153180895 @default.
• W1764744555 hasConcept C154945302 @default.
• W1764744555 hasConcept C155542232 @default.
• W1764744555 hasConcept C201995342 @default.
• W1764744555 hasConcept C22212356 @default.
• W1764744555 hasConcept C2776151529 @default.
• W1764744555 hasConcept C2776401178 @default.
• W1764744555 hasConcept C2777113093 @default.
• W1764744555 hasConcept C2780156472 @default.
• W1764744555 hasConcept C2780451532 @default.
• W1764744555 hasConcept C31972630 @default.
• W1764744555 hasConcept C34413123 @default.
• W1764744555 hasConcept C41008148 @default.
• W1764744555 hasConcept C41895202 @default.
• W1764744555 hasConcept C76155785 @default.
• W1764744555 hasConcept C90509273 @default.
• W1764744555 hasConcept C94915269 @default.
• W1764744555 hasConceptScore W1764744555C104114177 @default.
• W1764744555 hasConceptScore W1764744555C115961682 @default.
• W1764744555 hasConceptScore W1764744555C127413603 @default.
• W1764744555 hasConceptScore W1764744555C138885662 @default.
• W1764744555 hasConceptScore W1764744555C153180895 @default.
• W1764744555 hasConceptScore W1764744555C154945302 @default.
• W1764744555 hasConceptScore W1764744555C155542232 @default.
• W1764744555 hasConceptScore W1764744555C201995342 @default.
• W1764744555 hasConceptScore W1764744555C22212356 @default.
• W1764744555 hasConceptScore W1764744555C2776151529 @default.
• W1764744555 hasConceptScore W1764744555C2776401178 @default.
• W1764744555 hasConceptScore W1764744555C2777113093 @default.
• W1764744555 hasConceptScore W1764744555C2780156472 @default.
• W1764744555 hasConceptScore W1764744555C2780451532 @default.
• W1764744555 hasConceptScore W1764744555C31972630 @default.
• W1764744555 hasConceptScore W1764744555C34413123 @default.
• W1764744555 hasConceptScore W1764744555C41008148 @default.
• W1764744555 hasConceptScore W1764744555C41895202 @default.
• W1764744555 hasConceptScore W1764744555C76155785 @default.
• W1764744555 hasConceptScore W1764744555C90509273 @default.
• W1764744555 hasConceptScore W1764744555C94915269 @default.
• W1764744555 hasLocation W17647445551 @default.
• W1764744555 hasOpenAccess W1764744555 @default.
• W1764744555 hasPrimaryLocation W17647445551 @default.
• W1764744555 hasRelatedWork W127764718 @default.
• W1764744555 hasRelatedWork W1542939418 @default.
• W1764744555 hasRelatedWork W1552827497 @default.
• W1764744555 hasRelatedWork W2064021407 @default.
• W1764744555 hasRelatedWork W2154081650 @default.
• W1764744555 hasRelatedWork W2275815631 @default.
• W1764744555 hasRelatedWork W2336740878 @default.
• W1764744555 hasRelatedWork W2461914272 @default.
• W1764744555 hasRelatedWork W2563681188 @default.
• W1764744555 hasRelatedWork W2578968609 @default.
• W1764744555 hasRelatedWork W2591681541 @default.
• W1764744555 hasRelatedWork W2790263239 @default.
• W1764744555 hasRelatedWork W2890987528 @default.
• W1764744555 hasRelatedWork W2901505109 @default.
• W1764744555 hasRelatedWork W3130223753 @default.
• W1764744555 hasRelatedWork W3156313355 @default.
• W1764744555 hasRelatedWork W3206067408 @default.
• W1764744555 hasRelatedWork W601862425 @default.
• W1764744555 hasRelatedWork W209302195 @default.
• W1764744555 hasRelatedWork W2500958527 @default.
• W1764744555 isParatext "false" @default.
• W1764744555 isRetracted "false" @default.
• W1764744555 magId "1764744555" @default.
• W1764744555 workType "dissertation" @default.