Nagachetan Bangalore

An adaptive sample count particle filter

The particle filter technique has been used extensively over the past few years to track objects in challenging environments. Due to its nonlinear nature and the fact that it does not assume a Gaussian probability density function it tends to outperform other available tracking methods. A novel adaptive sample count particle filter (ASCPF) tracking method is presented in this paper for which the main motivation is to accurately track an object in crowded scenes using fewer particles and hence with reduced computational overhead. Instead of taking a fixed number of particles, a particle range technique is used where an upper and lower bound for the range is initially identified. Particles are made to switch between an active and inactive state within this identified range. The idea is to keep the number of active particles to a minimum and only to increase this as and when required. Active contours are also utilized to determine a precise area of support around the tracked object from which the color histograms used by the particle filter can be accurately calculated. This, together with the variable particle spread, allows a more accurate proposal distribution to be generated while using less computational resource. Experimental results show that the proposed method not only tracks the object with comparable accuracy to existing particle filter techniques but is up to five times faster.

Illumination invariant stationary object detection

A real-time system for the detection and tracking of moving objects that becomes stationary in a restricted zone. A new pixel classification method based on the segmentation history image is used to identify stationary objects in the scene. These objects are then tracked using a novel adaptive edge orientation-based tracking method. Experimental results have shown that the tracking technique gives more than a 95% detection success rate, even if objects are partially occluded. The tracking results, together with the historic edge maps, are analysed to remove objects that are no longer stationary or are falsely identified as foreground regions because of sudden changes in the illumination conditions. The technique has been tested on over 7 h of video recorded at different locations and time of day, both outdoors and indoors. The results obtained are compared with other available state-of-the-art methods.

Object tracking in a multi camera environment

Tracking objects in multi camera environments is an important requirement for video surveillance applications. A new active particle filter based tracking technique is presented, where objects are tracked across different cameras using a reduced number of particles. In order to cope with sudden colour and scale changes, a variable standard deviation value for spreading the particles is proposed. As the object moves from one scene to another, the number of particles along with the spread value is increased to minimize any effect of scale and colour change. The technique has been tested on live feeds from two different cameras and with scenes from the PETS dataset. The results have been compared with standard particle filtering techniques. It was found that not only did the proposed method result in almost similar tracking results but there is a 70% reduction in computational cost.

Automatic parameter adjustment of difference of Gaussian (DoG) filter to improve OT-MACH filter performance for target recognition applications

A wavelet-modified frequency domain Optimal Trade-off Maximum Average Correlation Height (OT-MACH) filter has been trained using 3D CAD models and tested on real target images acquired from a Forward Looking Infra Red (FLIR) sensor. The OT-MACH filter can be used to detect and discriminate predefined targets from a cluttered background. The FLIR sensor extends the filters ability by increasing the range of detection by exploiting the heat signature differences between the target and the background. A Difference of Gaussians (DoG) based wavelet filter has been use to improve the OT-MACH filter discrimination ability and distortion tolerance. Choosing the right standard deviation values of the two Gaussians comprising the filter is critical. In this paper we present a new technique for auto adjustment of the DoG filter parameters driven by the expected target size. Tests were carried on images acquired by the Apache AH-64 helicopter mounted FLIR sensor, results showing an overall improvement in the recognition of target objects present within the IR images.

Human detection using OT-MACH filter in cluttered FLIR imagery

An improvement to the Optimal Trade-off Maximum Average Correlation Height (OT-MACH) filter with the addition of a Rayleigh distribution filter has been used to detect humans in FLIR imagery scenes. The Rayleigh distribution filter is applied to the OT-MACH filter to provide a sharper low frequency cut-off which improves the OT-MACH filter performance in terms of target discrimination. The OT-MACH filter has been trained using a Computer Aided Design (CAD) model and tested on the corresponding real target object in high clutter environments acquired from a Forward Looking Infra Red (FLIR) sensor. Evaluation of the performance of the Rayleigh modified OT-MACH filter is reported for the recognition of humans present within the thermal infra-red image data set.

Determinant of homography-matrix-based multiple-object recognition

Finding a given object in an image or a sequence of frames is one of the fundamental computer vision challenges. Humans can recognize a multitude of objects with little effort despite scale, lighting and perspective changes. A robust computer vision based object recognition system is achievable only if a considerable tolerance to change in scale, rotation and light is achieved. Partial occlusion tolerance is also of paramount importance in order to achieve robust object recognition in real-time applications. In this paper, we propose an effective method for recognizing a given object from a class of trained objects in the presence of partial occlusions and considerable variance in scale, rotation and lighting conditions. The proposed method can also identify the absence of a given object from the class of trained objects. Unlike the conventional methods for object recognition based on the key feature matches between the training image and a test image, the proposed algorithm utilizes a statistical measure from the homography transform based resultant matrix to determine an object match. The magnitude of determinant of the homography matrix obtained by the homography transform between the test image and the set of training images is used as a criterion to recognize the object contained in the test image. The magnitude of the determinant of homography matrix is found to be very near to zero (i.e. less than 0.005) and ranges between 0.05 and 1, for the out-of-class object and in-class objects respectively. Hence, an out-of-class object can also be identified by using low threshold criteria on the magnitude of the determinant obtained. The proposed method has been extensively tested on a huge database of objects containing about 100 similar and difficult objects to give positive results for both out-of-class and in-class object recognition scenarios. The overall system performance has been documented to be about 95% accurate for a varied range of testing scenarios.

Improving OT-MACH filter performance for target recognition applications with the use of a Rayleigh distribution filter

An improvement to the wavelet-modified Optimal Trade-off Maximum Average Correlation Height (OT-MACH) filter with the use of the Rayleigh distribution filter is proposed. The Rayleigh distribution filter is applied to the OT-MACH filter to provide a sharper low frequency cut-off than the Laplacian of Gaussian based wavelet filter that has been previously reported to enhance OT-MACH filter performance. Filters are trained using a 3D CAD model and tested on the corresponding real target object in high clutter environments acquired from a Forward Looking Infra Red (FLIR) sensor. Comparative evaluation of the performance of the original, wavelet and Rayleigh modified OT-MACH filter is reported for the recognition of the target objects present within the thermal infra-red image data set.

Door surveillance using edge map-based Harris corner detector and active contour orientation

Accurately generating an alarm for a moving door is a precondition for tracking, recognizing and segmenting objects or people entering or exiting the door. The challenge of generating an alarm when a door event occurs is difficult when dealing with complex doors, moving cameras, objects moving or an obscured entrance of the door, together with the presence of varying illumination conditions such as a door-way light being switched on. In this paper, we propose an effective method of tracking the door motion using edge-map information contained within a localised region at the top of the door. The region is located where the top edge of the door displaces every time the door is opened or closed. The proposed algorithm uses the edge-map information to detect the moving corner in the small windowed area with the help of a Harris corner detector. The moving corner detected in the selected region gives an exact coordinate of the door corner in motion, thus helping in generating an alarm to signify that the door is being opened or closed. Additionally, due to the prior selection of the small region, the proposed method nullifies the adverse effects mentioned above and helps prevent different objects that move in front of the door affecting its efficient tracking. The proposed overall method also generates an alarm to signify whether the door was displaced to provide entry or exit. To do this, an active contour orientation is computed to estimate the direction of motion of objects in the door area when an event occurs. This information is used to distinguish between objects and entities entering or exiting the door. A Hough transform is applied on a specific region in the frame to detect a line, which is used to perform error correction to the selected windows. The detected line coordinates are used to nullify the effects of a moving camera platform, thus improving the robustness of the results. The developed algorithm has been tested on all the Door Zone video sequences contained with the United Kingdom Home Office i-LIDs dataset, with promising results.

Application of speed-enhanced spatial domain correlation filters for real-time security monitoring

A speed enhanced space variant correlation filer which has been designed to be invariant to change in orientation and scale of the target object but also to be spatially variant, i.e. the filter function becoming dependant on local clutter conditions within the image. The speed enhancement of the filter is due to the use of optimization techniques employing low-pass filtering to restrict kernel movement to be within regions of interest. The detection and subsequent identification capability of the two-stage process has been evaluated in highly cluttered backgrounds using both visible and thermal imagery acquired from civil and defense domains along with associated training data sets for target detection and classification. In this paper a series of tests have been conducted in multiple scenarios relating to situations that pose a security threat. Performance matrices comprised of peak-to-correlation energy (PCE) and peak-to-side lobe ratio (PSR) measurements of the correlation output have been calculated to allow the definition of a recognition criterion. The hardware implementation of the system has been discussed in terms of Field Programmable Gate Array (FPGA) chipsets with implementation bottle necks and their solution being considered.