Ajmal Shahbaz

Evaluation of background subtraction algorithms for video surveillance

This paper presents a comparative study of several state of the art background subtraction (BS) algorithms. The goal is to provide brief solid overview of the strengths and weaknesses of the most widely applied BS methods. Approaches ranging from simple background subtraction with global thresholding to more sophisticated statistical methods have been implemented and tested with ground truth. The interframe difference, approximate median filtering and Gaussian mixture models (GMM) methods are compared relative to their robustness, computational time, and memory requirement. The performance of the algorithms is tested in public datasets. Interframe difference and approximate median filtering are pretty fast, almost five times faster than GMM. Moreover, GMM occupies five times more memory than simpler methods. However, experimental results of GMM are more accurate than simple methods.

Probabilistic foreground detector for sterile zone monitoring

Detection of a moving object is often considered first step of multistage computer vision system such as visual surveillance. This paper proposes foreground detector based on Gaussian Mixture Models (GMM) for sterile zone monitoring. Each pixel is modeled by a mixture of Gaussians. Additionally, Morphological operations are incorporated on a foreground mask to reduce undesirable noise, thereby, restoring geometry of the detected object appreciably. The proposed method tested on i-LIDs dataset for sterile zone monitoring successfully detects and tracks foreground object in all video sequences.

Unified smoke and flame detection for intelligent surveillance system

This paper explains the way of unification of flame and smoke detection algorithms by merging the common steps into a single processing flow. Scenario, discussed in the current manuscript, considers using fixed surveillance cameras that allows using background subtraction to detect changes in a scene. Due to imperfection of background subtraction, foreground pixels, belonging to the same real object, are often separated. These pixels are united by morphological operations. All pixels are then labeled by connected components labeling algorithm, and tiny objects are removed since noticeable smoke and flames are to be detected. All the previous steps are processed only once, and then separate smoke and flame parts are started which use the same input image obtained after removing tiny objects. Smoke detection includes color probability, boundary roughness, edge density, and area variability filtering processes. Flame detection uses color probability, boundary roughness, and area variability filtering. Preliminary results show that applying unification to smoke and flame detection algorithms makes processing time similar to a single smoke detection algorithm if smoke and flame are processed in parallel. If the whole algorithm is implemented on a single thread, processing time is still lower comparing to running smoke and fire detection without unification. The result of unified processing part can also be used as input for multiple tasks of intelligent surveillance systems.

Probabilistic foreground detector with camouflage detection for sterile zone monitoring

Probabilistic modeling of background is extensively used for foreground detection in computer vision. Gaussian Mixture Models (GMM) is famous choice for detecting foreground in video sequences owing to ability of adapting background variation. However, GMM is prone to camouflage effect i.e. foreground object and background having same pixel intensity. This paper proposes foreground detector based on GMM with camouflage detection for sterile zone monitoring. Before modeling background, decision module based on third order image moments (skewness) is implemented to decide whether certain frame needs image enhancement. Then, histogram equalization (HE) is applied on such frame to differentiate between foreground object and background and then modeled using GMM. Morphological operations are incorporated on foreground mask to improve final results. The proposed method tested on i-LIDS dataset for sterile zone monitoring outperforms conventional GMM and detects camouflage object in all video sequences.

Estimation of walking direction for pedestrian path prediction from moving vehicle

This paper presents a method for estimating of walking direction for pedestrian path prediction. Pedestrian intending to laterally cross the street is observed by images which captured from a monocular camera mounted on the vehicle. The positional information of object is obtained by projecting the centroid of bounding box in the ground plane. Then, dependency between the real worlds, global coordinates and the position of object in the image is explained. The way to find the estimated distance of a pedestrian is introduced by using its position on the image. Walking direction of pedestrian is determined by concatenating consecutive frames. For pedestrian path prediction, Kalman filter (KF), interacting multiple models (IMM) and probabilistic hierarchical trajectory matching are evaluated. In experiments, publicly dataset is used. Distance of object (pedestrian), walking direction and accuracy of path prediction are evaluated. The largest distance error is 1.35 meters from the vehicle, walking direction accuracy is 97.50% and path prediction error is 0.08 meters.

Laser based collision warning system for high conflict vehicle-pedestrian zones

A collision risk estimation plays a crucial role in both driver and pedestrian safety in advanced driver assistance systems (ADAS) and autonomous vehicle navigation(AVN). In the proposed approach, an object warning collision system is implemented using a laser sensor. We focus on high conflict vehicle/pedestrian zones within a range of [20km/h, 30km/h]. The proposed method was implemented in four steps. Firstly, the lane region surface located ahead of the vehicle is detected. Secondly, a collision risk region is determined by using the vehicle required stopping distance. Thirdly, the objects within the lane surface region and the collision risk region are detected. Then for the all detected object, the centroid, radius information are extracted. Lastly, in order to prevent accidents as possible a collision risk estimation is implemented. Preliminary results were performed and tested on a group of consecutive frame captured while in motion to prove its effectiveness.

Estimation of collision risk for improving driver's safety

This paper introduces a method for analyzing the critical situation based on collision risk probability. Pedestrians in the scene are captured from a monocular camera mounted on the vehicle. Position information of object is extracted by projecting the centroid of bounding box to the ground plane. Five elements of collision criteria are used for our risk analysis. Pedestrian walking direction, its velocity and how aware pedestrian to the traffic are obtained from the pedestrian side. Car speed and relative distance of pedestrian from the car are extracted from car side. Then, with certain values of collision criteria, those elements are constructed. The critical situation is defined as joint probability of elements. Pedestrian are localized according to the critical situation as green for secure label, yellow for carefully and red for high priority to be alerted. A quantitative analysis is performed by measuring effectiveness of this approach. A real-world measurement and human perception survey are performed for evaluation. The performance evaluation shows our proposed method achieved average accuracy 87.5% and it significantly outperforms human perception survey with more than 30% improvement.

Dense optical flow in stabilized scenes for moving object detection from a moving camera

This paper proposes a method for detecting moving objects appeared in video captured by a moving camera. The proposed method relies on dense optical flow to differentiate moving objects from static background. Whenever video taken from a static camera is used, the dense optical flow itself is sufficient to determine the moving object in the scenes. However, in a non-static camera, all pixels are moving making which lead to incapability of optical flow to differentiate the moving objects from the static background. In order to solve this problem, a stabilization method is incorporated by the mean of global motion extraction, which can be done by analyzing the homography transformation between two consequtive frames. Finally, by applying a threshold on the dense optical flow, the region of moving object is acquired. The proposed method has been evaluated in the experiments and produce satisfying results with 98% accuracy.

Coarse-to-fine approach for fast correlation-based visual tracking

This paper proposes a coarse-to-fine approach for fast image tracking. The tracking method is built based on correlation tracker which employs online learning and fast detection by utilizing Fourier transform principles. Firstly, a small patch is extracted from a region near the tracked pixel. This patch is divided into a number of cells and then features are extracted from each cell, providing a set of training data. Together with the target values which are set as maximum at the center of the patch and getting smaller as the cell position getting farther from the center, a training is performed to determine the filter values. In the successive frame, the filter response is calculated to determine the position of the tracked pixel which is co-located with the maximum response of the filter. Since the features are extracted from cells, the new position of the tracked pixel is not precisely known. By employing a second detection at finer resolution within the corresponding cell, the ambiguity of the tracked pixel is eliminated. The proposed method was evaluated on a public dataset and the result shows that this strategy achieves a faster computation time compared to the baseline method.

Sterile zone monitoring with human verification

This paper proposes efficient real time method for sterile zone monitoring with human verification. The propose method consists of two main parts: Motion detection module and human verification module. The role of motion detection module is to segment out foreground object from background. Probabilistic Foreground Detector based on Gaussian Mixture Model(GMM) is used. Region of interest (ROI) obtained from motion detection module is fed into SVM classifier. SVM classifier is trained using HOG descriptor. The proposed method is tested on the standard datasets gives promising results.