Kaichi Fujimura

Pedestrian detection algorithm for on-board cameras of multi view angles

In this paper, a general algorithm for pedestrian detection by on-board monocular camera which can be applied to cameras of various view ranges in unified manner. The Spatio-Temporal MRF model extracts and tracks foreground objects as pedestrians and non-pedestrian distinguishing from background scenes as buildings by referring to motion difference. During the tracking sequences, cascaded HOG classifiers classify the foreground objects into the two classes of pedestrians and non-pedestrians. Before the classification, geometrical constraints on the relationship between heights and positions of the objects are examined to exclude the non-pedestrian objects. This pre-processing contributed to reducing the processing time of the classification while maintaining the classification accuracy. Due to the benefit of the tracking that the classifier can make decision totally considering Regions of Interest (ROIs) with same ID during consecutive images, this algorithm can operates quite robustly against noises and classification errors at each image frame.

Pedestrian Tracking Across Panning Camera Network

Video surveillance technology is getting important today, in order to maintain the safety of pedestrians passing through public spaces. Tracking pedestrians across the camera network is important to understand each pedestrian’s behavior from the image sequence of a long period. For that purpose, we developed an occlusion robust tracking algorithm of pedestrians in the panning images by the combination between the S-T MRF model and pattern recognition methods of Snakes and HOG classifier. Tracking in panning images would extend the field of view of single camera. In addition, we developed an algorithm to match pedestrians between cameras which have overlapping area with each other in their field of view. Finally, the tracking algorithm in panning images and the pedestrian matching algorithm between the overlapping images were combined to extend the area of pedestrian surveillance.

Pedestrian tracking through camera network for wide area surveillance

Tracking individual pedestrians through camera network is quite important for the security issue today. In the practical use, occlusion robust methods for pedestrian matching between cameras with overlapping view are necessary. However, such matching should be processed without camera calibrations that require real world coordination. In this paper, we proposed an occlusion robust method for pedestrian matching in overlapping view of contiguous cameras. Instead of real world coordination of cameras, the matching algorithm requires quite simple calibration between continuous cameras.

Pedestrian detection by on-board camera using collaboration of inter-layer algorithm

In this paper we present a robust pedestrian detection algorithm in low resolution on-board monocular camera image sequences of cluttered scenes. At first a motion based object detection algorithm is developed to detect foreground objects by analyzing horizontal motion vector. A cascade structure of rejection type classifier is utilized for our pedestrian detection system. Initial stage of cascade, simple rule based classification techniques are used to separate pedestrian from obvious road side structural object and later part of the cascade, a more complex algorithm which is a combination of Histogram of Oriented Gradients(HOG) and Support Vector Machine(SVM) based classification techniques are utilized to separate pedestrian from non-pedestrian objects. Finally, the image segments are tracked by our Spatio-Temporal Markov Random Field model(S-T MRF). Results show that our algorithms are promising for pedestrian detection in cluttered scenes.

On-board Pedestrian Detection by the Motion and the Cascaded Classifiers

This paper presents a general algorithm for pedestrian detection by on-board monocular camera, which can be applied to cameras of various viewing angle. The foreground objects which have discriminative motion-differences compared to that of background buildings are extracted as Regions of Interest (ROIs). Then, those ROIs are tracked by Spatio-Temporal MRF(S-T MRF) model as a possible pedestrian. Before classification process, some ROIs are rejected quickly by geometric constraints, calculated from coordinates between position of ROI and heights in real-world. Finally, ROIs are verified by HOG/Fisher cascade. The geometrical pre-processing saves the overall computational costs as well as decrease false positive rate. Also, algorithm compensates the detection rate due to temporal classification error in tracking cues, which improves the stability of tracking.

Behavior understanding at railway station by association of locational semantics and postures

The protection of critical transportation assets and infrastructure is an important topic in these days. In this paper, we develop a new rule based approach to smart video surveillance system for detecting situations where people may be in peril, as well as suspicious action or interactions at or near critical transportation assets. We analyze here three general types of human involved behaviors and interactions: (i) single pedestrian or no interaction, (ii) multiple pedestrian interactions, and (iii) pedestrian-facility/location interactions. The behavior analysis is accomplished through the development of geometric and motion visual features for each pedestrian. It is very simple and highly effective. The performance evaluation is carried out by using the video sequences taken in the real life environments of rail stations.

Behavior understandings based on state transition of postures and locations at railway station

Public safety and security in the railway station is significant. This paper describes a development of a framework composed of pedestrian detection and tracking, analysis based on both interactions between postures and location contexts and interactions between multiple pedestrians for behavior understanding in the railway station. In the railway station, a lot of pedestrians are passing through, and occlusions occur frequently. The occlusions have an influence on the accuracy of trajectories, and postures recognitions. Our framework uses ST-MRF model to reduce the influence of the occlusions. We also develop a method correcting the foot position of the occluded pedestrians to improve the accuracy of trajectories. In addition, the posture of occluded pedestrians can not be seen in single camera situation. For the problem, we develop a pedestrian matching technique between two cameras to detect the postures of the occluded pedestrians by selecting the camera without occlusions adaptively. By our framework, we improve the precision of behavior understandings in the railway station including occluded conditions.

Behavior understanding at railway station by postures and the pseud-trellis analysis of trajectories

Protection of critical transportation infrastructure such as railroad station is an important topic in these days. In this paper, we develop a new rule based approach to video surveillance system for detecting situations where pedestrians may be in peril, as well as suspicious action or interactions at railroad station. We analyze here two general types of pedestrian involved behaviors: (i) single pedestrian or no interaction, (ii) multiple pedestrian interactions. The behavior analysis is accomplished through the development of geometric and motion visual features for each pedestrian. The performance evaluation is carried out by using the video streams taken in the rail road stations. The results show that our algorithm is very simple and effective.

Pedestrian tracking across panning camera network

This paper describes developing of an occlusion robust tracking algorithm of pedestrians in the panning images by the combination between the S-T MRF model and pattern recognition methods of Snakes and HOG classifier. Tracking in panning images would extend the field of view of single camera. In addition, an algorithm to match pedestrians between cameras that have overlapping area with each other in their field of view is represented. Finally, the tracking algorithm in panning images and the pedestrian matching algorithm between the overlapping images are combined to extend the area of pedestrian surveillance.

Development of vision sensor for cooperative vehicle-highway systems

This paper describes a development of vision sensor that can detect shockwave propagation that is one of main factors of accidents in highway traffic flow. In addition, realization of a driver assistance system that informs arrival of such shockwave to drivers by the vision sensors is shown. To evaluate the reliability of the system, both recall rate and false rate of the system is investigated by sensing results of the sensors. As a result, it is shown that the detection of shockwaves and judgment of warning in the system should be decided in downstream sensors to minimalize these rates.