Location Classification of Detected Pedestrian

Joko Hariyono,Van-Dung Hoang,Kang-Hyun Jo 제어로봇시스템학회 2014 제어로봇시스템학회 국제학술대회 논문집 Vol.2014 No.10

This paper proposes a method to detect pedestrians from a single camera mounted on the vehicle then classify the location of the pedestrian to give information for the driver assistance system. The system consists of three stages. First, moving objects are detected using optical flows method. A moving object is extracted from the relative motion by segmenting the region representing the same optical flows after compensating the ego-motion of the camera. The regions of moving object are detected as transformed objects which are different from the previously registered background. Second, histogram of oriented gradients (HOG) features descriptor and linear support vector machine (SVM) are used to recognize the pedestrian from detected moving objects. Third, a heuristic method according to the image formation in advance from its geometrical coordinates is proposed. It is used for classify the location of the detected pedestrian using the region properties of the image. The image is classified into two regions, the road region in front of vehicle and the pedestrian movement region. The proposed method is evaluated using sequential images in outdoor environment, and the performance results shown the best pedestrian detection rate is 99.3% at 0.09 false positive rate. The location classification evaluation shown correct detection rate is 92.40%.

Moving Object Localization Using Optical Flow for Pedestrian Detection from a Moving Vehicle

Hariyono, Joko,Hoang, Van-Dung,Jo, Kang-Hyun Hindawi Publishing Corporation 2014 The Scientific World Journal Vol.2014 No.-

<P>This paper presents a pedestrian detection method from a moving vehicle using optical flows and histogram of oriented gradients (HOG). A moving object is extracted from the relative motion by segmenting the region representing the same optical flows after compensating the egomotion of the camera. To obtain the optical flow, two consecutive images are divided into grid cells 14 × 14 pixels; then each cell is tracked in the current frame to find corresponding cell in the next frame. Using at least three corresponding cells, affine transformation is performed according to each corresponding cell in the consecutive images, so that conformed optical flows are extracted. The regions of moving object are detected as transformed objects, which are different from the previously registered background. Morphological process is applied to get the candidate human regions. In order to recognize the object, the HOG features are extracted on the candidate region and classified using linear support vector machine (SVM). The HOG feature vectors are used as input of linear SVM to classify the given input into pedestrian/nonpedestrian. The proposed method was tested in a moving vehicle and also confirmed through experiments using pedestrian dataset. It shows a significant improvement compared with original HOG using ETHZ pedestrian dataset.</P>

Accuracy Enhancement of Omnidirectional Camera Calibration for Structure from Motion

Joko Hariyono,Wahyono,Kang-Hyun Jo 제어로봇시스템학회 2013 제어로봇시스템학회 국제학술대회 논문집 Vol.2013 No.10

This paper presents a technique to enhance accuracy of omnidirectional camera calibration which is applied structure from motion. We use two omnidirectional images of a predefined trihedron with checker board taken by a calibrate camera at an arbitrary location. Then, several point matches were picked manually from both views of the object. After that, compute and perform a 3D metric reconstruction of a real object from both images, by using the geometrical approach model estimated by single view calibration method. Then improved by the minimization error the distances among trihedral object use our proposed method. The overall experimental results show our proposed method gave smaller distance error.

Localization of Pedestrian Area from Hybrid Camera System

Joko Hariyono,Kang-Hyun Jo 제어로봇시스템학회 2015 제어로봇시스템학회 국제학술대회 논문집 Vol.2015 No.10

Accurate pedestrian path prediction and motion estimation are important tasks in the intelligent vehicle domain. The model incorporates the pedestrian pose recognition and lateral speed, motion direction and spatial layout of the environment. Pedestrian poses are recognized according to the spatial body language ratio. The region of interest of detected human is used. Then the centroid of the body relative to its width and height is used to define the pedestrian pose. Motion trajectory is obtained by using point tracking on that centroid, and then motion direction and estimated velocity are determined. Spatial layout is determined by the location of pedestrian with respect to road boundary. Input vector pedestrian pose, motion direction and speed, and the distance from the road lane are used by linear dynamic system. Instead of combining a number of subjects in a single model that will have to deal with the stylistic variations, all the subjects are separately trained in individual models. These models will be then hierarchically separated according to their action. Then, its classification will constrain the models to use for the prediction. Hybrid camera system is performed. A perspective camera is used for pedestrian localization with respect to car speed. And then an omnidirectional camera is used to enlarge field of view of monocular camera. Pedestrians are tracked in 360 degree around the vehicle. Experimental results show that the system has path predictions accuracy with the largest mean errors, 25 cm, for bending trajectories.

Analysis of Pedestrian Collision Risk using Fuzzy Inference Model

Joko Hariyono,Laksono Kurnianggoro,Wahyono,Kang-Hyun Jo 제어로봇시스템학회 2016 제어로봇시스템학회 국제학술대회 논문집 Vol.2016 No.10

The aim of this work is to improve driver awareness by proposing a collision risk analysis method. Pedestrian in the scene is observed by sequential frames from monocular camera mounted on the car. Positional information of object is extracted by projecting the centroid of bounding box on the ground plane. Four elements of collision criteria are constructed which are pedestrian walking direction, its velocity, car speed and relative distance of pedestrian. The analysis of collision risk is performed using fuzzy inference method that is used for calculating the degree of risk. Furthermore, localization of pedestrian is performed according to its risk score. The pedestrian with low collision score is labeled as low risk (green), pedestrian which is increasing its collision score is considered as medium risk (yellow) and pedestrian with high collision score is labelled as high risk (red). A quantitative analysis is performed by measuring effectiveness of this approach. The performance evaluation shows our proposed method achieved average accuracy 87.5% and it significantly outperforms human perception with more than 25% improvement.

Detection of pedestrian crossing road: A study on pedestrian pose recognition

Hariyono, Joko,Jo, Kang-Hyun Elsevier BV 2017 Neurocomputing Vol. No.

<P><B>Abstract</B></P> <P>Detection of pedestrian crossing road is the objective of this work. The model incorporates the pedestrian pose recognition and lateral speed, motion direction and spatial layout of the environment. Pedestrian poses are recognized according to the spatial body language ratio. The center of mass of the body relative to its width and height is used to define the pedestrian pose. Motion trajectory is obtained by using point tracking on the centroid of detected human region, and then estimated velocity is determined. Spatial layout is determined by the distance of the pedestrian to the road lane boundary. These models will be then hierarchically separated according to their action (walking, starting, bending and stopping). In order to classify the pedestrian crossing road, a walking human model is proposed. A walking human is defined by ratio of the centroid location from the ground plane divided by the height of bounding box that should satisfy a constraint. The proposed algorithms are evaluated using publicly available datasets and our pedestrian walking dataset. The performance result shows that the correct pedestrian crossing road classification is 98.10%.</P>

Body part boosting model for carried baggage detection and classification

Wahyono,Hariyono, Joko,Jo, Kang-Hyun Elsevier 2017 Neurocomputing Vol.228 No.-

<P><B>Abstract</B></P> <P>In the automatic video surveillance system, the detection of a human carrying baggage is a potentially important objective for security and monitoring purposes in the public spaces. This paper introduces a new approach for detecting and classifying baggage carried by a human on the images. It utilizes the spatial information of the baggage in reference to the body of the human carrying it. A human-baggage detector is modeled by the body parts of a human, including the head, torso, leg, and baggage parts. The feature descriptors are extracted for each part based on its characteristics and these features are further trained using a support vector machine (SVM) classifier. A mixture model is built specifically for the baggage part due to a significant variation in shape, size, color, and texture. The boosting strategy constructs a strong classifier by combining a set of weak classifiers which are obtained by training the body part. The proposed method has been extensively evaluated using the public datasets. The experimental results confirm that the proposed method is viable for a state-of-the-art in the carried baggage detection and classification system.</P>

Adaptive Frame Rate Streaming Strategy for Intelligent Surveillance System

Alexander Filonenko,Wahyono,Joko Hariyono,Ajmal Shahbaz,Youl-Kyeoung Lee,Kang-Hyun Jo 제어로봇시스템학회 2016 제어로봇시스템학회 국내학술대회 논문집 Vol.2016 No.3