Yupin Luo

Real-Time Pedestrian Detection and Tracking at Nighttime for Driver-Assistance Systems

Pedestrian detection is one of the most important components in driver-assistance systems. In this paper, we propose a monocular vision system for real-time pedestrian detection and tracking during nighttime driving with a near-infrared (NIR) camera. Three modules (region-of-interest (ROI) generation, object classification, and tracking) are integrated in a cascade, and each utilizes complementary visual features to distinguish the objects from the cluttered background in the range of 20-80 m. Based on the common fact that the objects appear brighter than the nearby background in nighttime NIR images, efficient ROI generation is done based on the dual-threshold segmentation algorithm. As there is large intraclass variability in the pedestrian class, a tree-structured, two-stage detector is proposed to tackle the problem through training separate classifiers on disjoint subsets of different image sizes and arranging the classifiers based on Haar-like and histogram-of-oriented-gradients (HOG) features in a coarse-to-fine manner. To suppress the false alarms and fill the detection gaps, template-matching-based tracking is adopted, and multiframe validation is used to obtain the final results. Results from extensive tests on both urban and suburban videos indicate that the algorithm can produce a detection rate of more than 90% at the cost of about 10 false alarms/h and perform as fast as the frame rate (30 frames/s) on a Pentium IV 3.0-GHz personal computer, which also demonstrates that the proposed system is feasible for practical applications and enjoys the advantage of low implementation cost.

Rapid and brief communication: Camera calibration with one-dimensional objects moving under gravity

Camera calibration using one-dimensional (1D) rigid objects is arresting the attentions of researchers since the easy-to-construct geometrical structure of the apparatuses. In this paper, we extend the motion patterns applicable for calibration with the motion of 1D objects. We show that a 1D object with three or more markers, rotating around one marker which is moving in a plane, provides constraint equations on camera intrinsic parameters. A stick moving under gravity without other forces acting on performs such a motion. Simulated tests show the feasibility and numerical robustness of this method.

Weather Recognition Based on Images Captured by Vision System in Vehicle

Weather recognition is widely required in many areas, and it is also a challenging and brand-new subject. This paper proposes an approach to recognize weather based on images captured by in-vehicle vision system. We bring three groups of features, including histogram of gradient amplitude, HSV color histogram, road information, and employ an algorithm based on Real AdaBoost, making use of the category structure to achieve the task of classification. Experiments confirm superior performances on our dataset collected from images captured by vision system.

Constraints on general motions for camera calibration with one-dimensional objects

This paper focuses on two problems in camera calibration with one-dimensional (1D) objects: (a) to find out the general motion patterns well suited for solving the calibration problem, and (b) to improve the robustness and accuracy of the method. Firstly, a sufficient and necessary condition for the solvability of 1D calibration with general motions is proved. Then the special motion of tossing a 1D object is provided as an example to illustrate the correctness and feasibility of this condition. After that some practical issues on obtaining the solution are inspected. By avoiding singularities, the precision and robustness of the method are improved: the relative mean errors are reduced to less than 5% at the noise level of one pixel which surpasses the state-of-the-art methods of the same category.

Attention recognition of drivers based on head pose estimation

Recognition of driverpsilas attention plays an important role in driver assistance system, which can recognize the driverpsilas state and prevent the driver from distracting his attention. To some extent, the driverpsilas head pose implies his attentionpsilas orientation. In this paper we describe a vision-based scheme of estimating the driverpsilas head pose, which is based on estimating the relative pose between adjacent views. Scale-Invariant Feature Transform (SIFT) descriptors are utilized to match the corresponding feature points between two adjacent views. We also bring forward a method to capture the key frames online through fusing the smoothness assumption of headpsilas motion and other prior motion information, in order to eliminate the accumulating errors during continuous pose accumulation. The appropriate key frames will be chosen as for the different frames within different pose range to refresh the result of pose estimation, which can reduce the accumulating error during the process of head pose estimation greatly. The experiment shows that our system can work very well, even within large rotation scope and over long video sequences.

Stagnation Analysis in Particle Swarm Optimization

Particle swarm optimization (PSO) has shown to be an efficient, robust and simple optimization algorithm, and has been successfully applied to many different kinds of problems. But it is still an open problem that why PSO can be successful. Most of current PSO studies are empirical, with only a few theoretical analyses, and these theoretical studies concentrate mainly on simplified PSO systems, discarding randomness. In order to improve the understanding of real stochastic PSO algorithm, this paper presents a formal stochastic analysis of the stochastic PSO algorithm, which involves with randomness. The stochastic properties of particle trajectories in stagnation phase are studied in details

Single-image superresolution based on local regression and nonlocal self-similarity

Abstract. The challenge of learning-based superresolution (SR) is to predict the relationships between low-resolution (LR) patches and their corresponding high-resolution (HR) patches. By learning such relationships from external training images, the existing learning-based SR approaches are often affected by the relevance between the training data and the LR input image. Therefore, we propose a single-image SR method that learns the LR-HR relations from the given LR image itself instead of any external images. Both the local regression model and nonlocal patch redundancy are exploited in the proposed method. The local regression model is employed to derive the mapping functions between self-LR-HR example patches, and the nonlocal self-similarity gives rise to a high-order derivative estimation of the derived mapping function. Moreover, to fully exploit the multiscale similarities inside the LR input image, we accumulate the previous reconstruction results and their corresponding LR versions as additional example patches for the subsequent estimation process, and adopt a gradual magnification scheme to achieve the desired zooming size step by step. Extensive experiments on benchmark images have validated the effectiveness of the proposed method. Compared to other state-of-the-art SR approaches, the proposed method provides photorealistic HR images with sharp edges.

A multi-stage classifier based algorithm of pedestrian detection in night with a near infrared camera in a moving car

The paper presents an algorithm of pedestrian detection in night using a near infrared camera which is installed in a moving car. To deal with several kinds of primary disturbances in the road, we adopt a multi-stage classifier which eliminates certain kinds of disturbances in every stage. In the final stage, a classifier based on 2D pedestrian shape model is especially designed to make decision. The speed of detection can be accelerated due to the multi-stage strategy. The experiment shows our method is promising.

Making full use of spatial-temporal interest points: An AdaBoost approach for action recognition

Although spatial-temporal interest points (STIPs) with bag of words strategy have achieved success in action recognition, they lose much information during forming histograms, especially the relations among STIPs. We propose to use effective human body regions (EHBRs) to find these relations in order to compensate for bag of spatial-temporal words (BOW). Combining bag of spatial-temporal words and EHBRs, the AdaBoost approach is used to achieve high accuracy. Experiments on benchmark dataset KTH verify our approach effectiveness and efficiency.

Nighttime Pedestrian Detection with Near Infrared using Cascaded Classifiers

This paper presents a novel nighttime pedestrian detection approach only using a near infrared camera, which can be used in a practical driver assistance systems. This method can be divided into three steps: selection step, preprocess step and recognition step. Firstly, objects in the video are separated with an adaptive dual thresholds segmentation method in the selection step; Secondly, most of non-pedestrians are discarded with some constraints in the preprocess step; Finally, in the recognition step a cascaded classifiers with Histograms of Oriented Gradients and Adaptive Boosting Algorithm are introduced. Experiments on video sequences show that the proposed pedestrian detection approach has a high detection rate as well as a very low false alarm rate and run in real-time.