Yanwu Xu

An Efficient Tree Classifier Ensemble-Based Approach for Pedestrian Detection

Classification-based pedestrian detection systems (PDSs) are currently a hot research topic in the field of intelligent transportation. A PDS detects pedestrians in real time on moving vehicles. A practical PDS demands not only high detection accuracy but also high detection speed. However, most of the existing classification-based approaches mainly seek for high detection accuracy, while the detection speed is not purposely optimized for practical application. At the same time, the performance, particularly the speed, is primarily tuned based on experiments without theoretical foundations, leading to a long training procedure. This paper starts with measuring and optimizing detection speed, and then a practical classification-based pedestrian detection solution with high detection speed and training speed is described. First, an extended classification/detection speed metric, named feature-per-object (fpo), is proposed to measure the detection speed independently from execution. Then, an fpo minimization model with accuracy constraints is formulated based on a tree classifier ensemble, where the minimum fpo can guarantee the highest detection speed. Finally, the minimization problem is solved efficiently by using nonlinear fitting based on radial basis function neural networks. In addition, the optimal solution is directly used to instruct classifier training; thus, the training speed could be accelerated greatly. Therefore, a rapid and accurate classification-based detection technique is proposed for the PDS. Experimental results on urban traffic videos show that the proposed method has a high detection speed with an acceptable detection rate and a false-alarm rate for onboard detection; moreover, the training procedure is also very fast.

Airborne moving vehicle detection for video surveillance of urban traffic

Urban traffic surveillance, which is designed to improve traffic management, is an important part of intelligent traffic system (ITS). In particular, airborne moving vehicle detection has become a new but hot research area since its wide view and low cost. However, airborne urban traffic surveillance is impacted by many difficulties such as camera vibration, vehicle congestion, background variance, serious thermal noise etc. Therefore, image subtraction and thermal image processing have low detection rate, while the optical flow method cannot meet the real-time application. In this paper, we propose a coarse-to-fine method, which can be divided into two stages of pre-processing and classification inspection. In pre-processing stage, the candidates regions of moving vehicle are obtained by employing Road Detection, Removal of Non-vehicle Regions and Moving Regions Extraction. The speed of this stage is fast but there is still relatively high false-positive-rate. In classification inspection stage, a well-trained cascade classifier, which refines the candidate regions, is designed to maintain a higher detection rate and a lower false alarm rate. Experimental results demonstrate that compared with representative algorithms, our method reach better performance in detection rate and false-positive-rate, while meeting the needs of real-time application.

Airborne moving vehicle detection for urban traffic surveillance

At present, moving vehicle detection on airborne platform has been an important technology for urban traffic surveillance. In such a situation, most commonly used methods (e.g. image subtraction) could hardly work well because of some additional difficulties such as slow movement of vehicles and jam. This paper proposed a new moving vehicle detection method named MVD-RD for airborne urban traffic surveillance. First, the non-road regions are extracted using road detection technique. Secondly, the non-road regions with no vehicles are removed according to their size. As a result of this two-stage regions shrinkage, the detection area reduces a lot. Finally, to the reduced area, image subtraction is used to get all moving regions and then moving vehicles can be accurately filtered in a simple way. The experimental results show that, compared with traditional image subtraction methods used in airborne moving vehicle detection, the proposed MVD-RD method achieves much better performance in detection rate, false alarm rate, and detection speed.

A Low-Cost Pedestrian Detection System with a Single Optical Camera

This paper presents a low-cost solution for pedestrian detection using a single optical video camera on a moving vehicle. Since only one optical camera can collect just a little original information, our system scan two sequential frames to get both appearance and motion information. We use zoom image and slide window techniques to select the objective region, apply a cascaded classifier combined with statistical learning and SVM to recognize human body, adopt zoom-scale to estimate the distance from a pedestrian and develop a distance transform algorithm to forecast his/her orientation. This system is suitable for detecting pedestrians in the range of 0.3-20 meters in the city traffic with the speed under 50 km/h. The test with videos of real city traffic indicates that our system has got acceptable detecting rate and processing speed

An effective crossing cyclist detection on a moving vehicle

Vision based cyclist detection is a new application in the field of intelligent transportation. Compared with pedestrian detection, this new problem is more challenging because various appearence and motion of bicycles increase the diversity of the detection objects; therefore existing pedestrian detection approaches can hardly get good overall performance because cyclist detection requires more information represented by more effective features to enable detection. For general object detection and pedestrian detection, histogram of oriented gradient (HOG) features achieved great success; however it have two major drawbacks: time-consuming caused by dense/overlap sampling and only local information is retained. In this paper, we proposed a more effective feature extraction method (i.e., HOG-LP) to overcome the drawbacks of general HOG feature extraction for crossing cyclist detection. On one hand, an improved light/non-overlap sampling method is proposed to speed up HOG feature extraction; on the other hand, pyramid sampling is utilized to extract additional global features in different scale spaces in order to retain more information for high classification accuray. With efficient feature extraction, a linear SVM classifier is used to further increase the detection speed. The experimental results tested on urban traffic videos show the effectiveness of the proposed method on crossing cyclist detection.

Extended Kalman filter based pedestrian localization for collision avoidance

The practical driving safety assistant system should be able to estimate the possibility of pedestrian-vehicle collision, which includes pedestrian detection and localization as well as collision prediction. Until now, many works concentrated on pedestrian detection and achieved some progress. For collision prediction, it is essential to locate the pedestrian precisely; however, the localization problem still needs to be further studied. At present, most researches adopted expensive equipments (e.g. millimeter wave radar and laser scanner) to run away from the difficulties; and many others used multi-cameras to solve this problem. In our previous work, we proposed a low-cost pedestrian detection system with a single optical camera, which performanced well in pedestrian detection. Basing on the detection system, an extended Kalman filter based pedestrian localization model/methodology is proposed in this paper. The localization model sets up proper relation between state vector and observation vector and chooses proper initial state for the Kalman filter using perspective projection principle, which guarantees the proposed filter to estimate the location of pedestrian quickly and actually. The experimental results have validated that the accuracy of the proposed localization model/methodology may meet the requirements of a practical collision avoidance system.

Pedestrian Detection with Local Feature Assistant

Until now, existing pedestrian detection systems usually use global features (e.g. appearance or motion) of human body to detect pedestrian; however, the detection rate needs to be improved in many situations since sometimes the global features can not be obtained. For example, a pedestrian may be partly covered by a car or his/her part may hide into the background. Therefore it is essential to adopt some local features of key parts of human body to assist pedestrian detection. In this paper, we propose a method using some key local features of human body to help pedestrian detection. Since the introduction of additional features will cost the system more time, in order to ensure the detection speed, we firstly use both appearance and motion global features of human body to select candidates, and then use local features of head and leg to do further confirmation. In the confirmation stage, we use three kinds of local features (head appearance, face color and hair color) to detect the head of each candidate; at the same time, we also choose some particular local appearance features to detect the leg. The experimental results indicate that this method can improve detection rate with almost the same detection speed; additionally, it can reduce false alarm sometimes.

An optimized hierarchical classifier for pedestrian detection

Classification is an essential technology in Pedestrian Detection System (PDS). Until now, single-classifier and basic cascaded classifier had been widely used in PDS; however, most of them can hardly satisfy the 3 requirements at the same time: high detection speed, high detection rate and low false positive rate. In this paper, we proposed an optimized hierarchical classifier which can satisfy the 3 requirements. The proposed method adopted corse-to-fine and early-rejection principles to achieve global high performance. It consists of two hierarchies, the first one is used to quickly reject non-pedestrian objects and select out only a few candidates; the second one makes further verification to these candidates. Furthermore, each hierarchy was optimized with statistical models basing on experiments; and each hierarchy is a treelike classifier which has specific optimization demands. At last, an overall performance evaluation standard is proposed, and the experimental results showed that the proposed classifier had better overall performance.

Feature subset selection based on co-evolution for pedestrian detection

An appropriate subset of features is needed for a classification-based pedestrian detection system since its performance is greatly affected by the features adopted. Moreover, the combination of different types of features (eg, grey-scale, colour) could improve the detection accuracy, so it is helpful to obtain a feature subset and the proportion of each type simultaneously for the classifier. However, because a larger number and various types of features are generally extracted to represent pedestrians better, it is difficult to achieve this. This paper proposed a co-evolutionary method to solve this problem. In the feature subset selection method, each sub-population mapped to one type of pedestrian feature, and then all sub-populations evolved co-operatively to obtain an optimal feature subset. Moreover, a strategy was specially designed to adjust the sub-population size adaptively in order to improve the optimizing performance. The proposed method has been tested on pedestrian detection applications and the experimental results illustrate its better performance compared with other methods such as genetic algorithm and AdaBoost.

Co-Evolution based Feature Selection for Pedestrian Detection

In a pedestrian detection system, the most critical requirement is to quickly and reliably determine whether a candidate region contains a pedestrian. The detection ability of whole system determines directly upon quality of chosen features. However, due to the large number and various types of available features, it is difficult to find an optimal feature subset and acquire the proper feature proportion at the same time for most traditional methods including AdaBoost Algorithm. This paper presents a co-evolutionary method with sub-population size adjusting strategy for the feature selection problem in pedestrian detection system. Our method is able to find an optimal feature subset and adjust feature proportion to a proper state in the mean time. Experiments show that our method performs better than AdaBoost Algorithm.