Visual object detection with deformable part models

We describe a general method for building cascade classifiers from part-based deformable models such as pictorial structures. We focus primarily on the case of star-structured models and show how a simple algorithm based on partial hypothesis pruning can speed up object detection by more than one order of magnitude without sacrificing detection accuracy. In our algorithm, partial hypotheses are pruned with a sequence of thresholds. In analogy to probably approximately correct (PAC) learning, we introduce the notion of probably approximately admissible (PAA) thresholds. Such thresholds provide theoretical guarantees on the performance of the cascade method and can be computed from a small sample of positive examples. Finally, we outline a cascade detection algorithm for a general class of models defined by a grammar formalism. This class includes not only tree-structured pictorial structures but also richer models that can represent each part recursively as a mixture of other parts.

Automatic object detection is a rapidly evolving area in surveillance and autonomous vehicles. Deformable part model (DPM) is a well-known object detector for its high precision and speed bottleneck. This paper proposes a very fast object detection pipeline based on complementary techniques to accelerate DPM. A recent fast feature pyramid technique is employed with look-up table HOG features, Fast Fourier Transform and early classification technique to speed up DPM and maintain its accuracy. We exploit SIMD optimization and multiple cores to achieve a real time detector. Our results shows that we achieve a speed-up of 50x and 6x on a single core over DPM and cascade DPM respectively. Our optimized version processes a 640×480 pixel image at 38 fps.

In complex scene, considering traditional object detection methods based on feature points have exposed many problems, such as undetected points, low detected ratio and cannot well process object occlusion and scaling situation, this paper proposes a detection method which based on a deformable part model. The method uses histogram of oriented gradient (HOG) feature as the object description, and the deformable part model includes a global template and several high-resolution templates. And the method uses the support vector machine (SVM) training the object model. In the learning process, after the HOG feature extracted, the method modifies the HOG feature, and then uses the principal component analysis (PCA) method reducing feature dimensions to avoid over-learning, and improve the detection rate in the detection process. The experiment results shows that the method proposed can better process object occlusion or scaled situation, and there’s also an improvement in detection ratio.

We describe an adaptive weighted deformable part model for object detection based on traditional deformable part model(DPM). In the original DPM model, we find that the high response score region calculated by the template filter as high-energy regions, which indicates that the influence on the detection results is greater. The parts can affect the results of object detection, some important parts may directly determine the accuracy of the results, and some unimportant parts even produce bad impacts. To reduce the adverse effects caused by unimportant part filter, we add an adaptive coefficient strategy to the traditional method, which could improve the accuracy of object detection without efficiency loss. The proposed algorithm is better in accuracy compared with the traditional deformable part model, especially in the case of occlusion, with the same efficiency.

Pedestrian detection plays an important role in unmanned technology. Because of the high efficiency and robustness of the deformable part model, it has been widely applied to the field of pedestrian detection. At present, how to effectively reduce the risk of partial screening of pedestrians has been based on the pattern recognition of pedestrian detection technology in the hot spots. Aiming at this problem, after analyzing the deformable part model deeply, this paper creatively proposes a pedestrian detection method with improved deformable part model. By training the two-pedestrian deformable part model, the method is adopted to reduce the pedestrian detection in the pedestrian detection by matching the image sub-region and matching the matching result. It is shown that the method can improve the detection efficiency while ensuring the detection efficiency, while ensuring the effectiveness of the whole algorithm to meet the real-time requirements of unmanned technology.

Object detection is fundamental for realization of self-driving cars and smart traffic management. Various vision based techniques are based on Deformable part models (DPMs) and recently convolutional neural network (CNN) for its anticipated high precision. The DeepPyramid DPM model maps DPM into multiple CNN layers as an assembly of the two concepts with intensive computations. While the recent Region Proposal Network (RPN) allows convolutional layers sharing with further detection steps to accelerate Fast R-CNN. Our framework employs these cost-free RPN proposals and DenseNet with Fast Fourier Transform (FFT) to introduce a faster and more accurate class independent and region based DeepPyramid DPM detector. Our evaluation on four different benchmarks shows that our technique processes PASCAL VOC at 20 fps and VGA images at 14 fps on a GPU (about 12x faster than DeepPyramid DPM with better precision).

Part-based models have become the mainstream approach for visual object classification and detection. The key tools adopted by the most methods are interest point detectors and descriptors, shared codes for object parts (visual codebook) and discriminative learning using positive and negative class examples. Distinction of our method from the existing part-based methods for object detection is the use of sparse class-specific landmarks with semantic meaning. The landmarks are the additional distinguished information of object location in the proposed framework. Additionally, localising semantic and discriminative landmarks (object parts) is significant in other related applications of computer vision, such as facial expression recognition and pose/orientation estimation of objects. Therefore, we propose a model which deviates from the mainstream by the fact that the object parts' appearance and spatial variation, constellation, are explicitly modelled in a generative probabilistic manner. With using only positive examples our method can achieve object detection accuracy comparable to state-of-the-art discriminative method.

Tracking of previously unseen, articulated objects is an active research area. Recently, Deformable Parts Models (DPMs) have been used to improve the online tracking performance for bounding-box trackers. In this paper, we extend the DPM with global priors which enforce consistency with foreground/background segmentation cues. We propose a Dual Decomposition approach and show how to efficiently solve the high-order coupling constraints as a feasible sub-problem. The proposed approach is evaluated on the VOT online tracking benchmark, outperforming the baseline in both tracking accuracy and robustness. We further show that in presence of stable image segmentation cues, the flexibility of a generic DPM generated from a single reference frame can be improved by introducing the concept of part visibility, the visibility-aware DPM (VDPM). This allows for fine-grained articulated object tracking using an automatically generated DPM from a single template image.

Vehicle detection at night time is a challenging problem due to low visibility and light distortion caused by motion and illumination in urban environments. This paper presents a method based on the deformable object model for detecting and classifying vehicles by using monocular infra-red cameras. As some features of vehicles, such as headlight and taillights are more visible at night time, we propose a weighted version of the deformable part model. We define weights for different features in the deformable part model of the vehicle and try to learn the weights through an enormous number of positive and negative samples. Experimental results prove the effectiveness of the algorithm for detecting close and medium range vehicles in urban scenes at night time.

Deformable part models (DPM) already proved great performance for objects detection and many extensions have already published in literature. DPMs generally have high performance, though they dramatically fail to detect objects in challenging environments such as night time. This paper proposes a method based on the idea of latent parts to optimize the structure of objects in deformable part models. Even in challenging environment some parts of the objects are still visible and latent parts can optimize the structure of DPM's object model to catch significant features. We have evaluated proposed method to detect cars at night time in urban area using IR camera. It is a challenging problem due to low visibility, light distortion and illumination/glare in urban area. Experimental results prove the effectiveness of the model to detect close and medium range cars in urban scenes at night time.

Density-aware generative algorithms learning from positive examples have verified high recall for visual object detection, but such generative methods suffer from excessive false positives which leads to low precision. Inspired by the recent success of detection-recognition pipeline with deep neural networks, this paper proposes a two-step framework by training a generative detector with positive samples first and then utilising a discriminative model to get rid of false positives in those detected bounding box candidates by the generative detector. Evidently, the discriminative model can be viewed as a post-processing step which improves the robustness by distinguishing true positives from false positives that confuse the generative detector. We exemplify the proposed approach on public ImageNet classes to demonstrate the significant improvement on precision while using only positive examples in training.

In this paper, we address the problems of contour detection, bottom-up grouping, object detection and semantic segmentation on RGB-D data. We focus on the challenging setting of cluttered indoor scenes, and evaluate our approach on the recently introduced NYU-Depth V2 (NYUD2) dataset (Silberman et al., ECCV, 2012). We propose algorithms for object boundary detection and hierarchical segmentation that generalize the $$gPb-ucm$$gPb-ucm approach of Arbelaez et al. (TPAMI, 2011) by making effective use of depth information. We show that our system can label each contour with its type (depth, normal or albedo). We also propose a generic method for long-range amodal completion of surfaces and show its effectiveness in grouping. We train RGB-D object detectors by analyzing and computing histogram of oriented gradients on the depth image and using them with deformable part models (Felzenszwalb et al., TPAMI, 2010). We observe that this simple strategy for training object detectors significantly outperforms more complicated models in the literature. We then turn to the problem of semantic segmentation for which we propose an approach that classifies superpixels into the dominant object categories in the NYUD2 dataset. We design generic and class-specific features to encode the appearance and geometry of objects. We also show that additional features computed from RGB-D object detectors and scene classifiers further improves semantic segmentation accuracy. In all of these tasks, we report significant improvements over the state-of-the-art.

In this paper we describe a detector of facial landmarks based on the Deformable Part Models. We treat the task of landmark detection as an instance of the structured output classification problem. We propose to learn the parameters of the detector from data by the Structured Output Support Vector Machines algorithm. In contrast to the previous works, the objective function of the learning algorithm is directly related to the performance of the resulting detector which is controlled by a user-defined loss function. The resulting detector is real-time on a standard PC, simple to implement and it can be easily modified for detection of a different set of landmarks. We evaluate performance of the proposed landmark detector on a challenging “Labeled Faces in the Wild” (LFW) database. The empirical results demonstrate that the proposed detector is consistently more accurate than two public domain implementations based on the Active Appearance Models and the Deformable Part Models. We provide an open-source implementation of the proposed detector and the manual annotation of the facial landmarks for all images in the LFW database.

The video monitoring system is susceptible to illumination and other factors. To effectively solve the problem of high-precision detection of truck face in complex highway scenes, a truck face detection method based on deformable parts model is proposed and compared with four other truck face detection methods based on machine learning in this paper. Based on the vehicle image database of Southeast University, comparative experiments are carried out and the theoretical analysis and experimental results show that the performance of the truck face detection method based on deformable parts model is better than the other four truck face detection methods based on machine learning, and its verification accuracy is better than 98%.

Automatic weapon platform is one of the important research directions at domestic and overseas, it needs to accomplish fast searching for the object to be shot under complex background. Therefore, fast detection for given target is the foundation of further task. Considering that chest-shape target is common target of shoot practice, this paper treats chestshape target as the target and studies target automatic detection method based on Deformable Part Models. The algorithm computes Histograms of Oriented Gradient(HOG) features of the target and trains a model using Latent variable Support Vector Machine(SVM); In this model, target image is divided into several parts then we can obtain foot filter and part filters; Finally, the algorithm detects the target at the HOG features pyramid with method of sliding window. The running time of extracting HOG pyramid with lookup table can be shorten by 36%. The result indicates that this algorithm can detect the chest-shape target in natural environments indoors or outdoors. The true positive rate of detection reaches 76% with many hard samples, and the false positive rate approaches 0. Running on a PC (Intel(R)Core(TM) i5-4200H CPU) with C++ language, the detection time of images with the resolution of 640 × 480 is 2.093s. According to TI company run library about image pyramid and convolution for DM642 and other hardware, our detection algorithm is expected to be implemented on hardware platform, and it has application prospect in actual system.