Pranav Mantini
P. Mantini, S. Shah. A Signal Detection Theory Approach for Camera Tamper Detection. pdf
Abstract: Camera tamper detection is the ability to detect faults and operational failures in video surveillance cameras by analyzing the video. Researchers have increasingly focused on such techniques attributing to the ubiquitous deployment of large scale surveillance systems. In this paper a signal detection theory approach is proposed to quantitatively analyze the information being captured by the camera and to detect tampers. Signal activity is used as a feature to measure the amount of information in the image. Normal operating parameters of a camera are modeled as a Gaussian mixture model (GMM) that is trained over synthetic training data. This is employed for tamper detection using signal activity as features. To reduce the effects of noise, a Kalman filter is used to estimate the signal activity of the images. Experimental results show that the proposed approach out performed the state-of-the-art in detecting tampered images with higher accuracy while generating lower false alarms.
P. Mantini, S. Shah. Multiple people tracking using contextual trajectory forecasting. pdf
Abstract: People tracking is the ability to identify the position of a specified person in the camera view with the progression of time. Trajectory forecasting is the task of predicting the likely path that a person might take to reach a destination. Contextual trajectory forecasting (CTF) leverages the 3D geometric information and static objects in the environment along with observed behavioral norms for human path prediction. In this paper, we enhance CTF to also account for dynamic objects in the environment (like other humans) for prediction. The proposed tracking algorithm makes use of traditional HSV histogram appearance features for detection and combines it with the enhanced CTF for tracking. A maximum likelihood minimum mean square error data association filter is used to probabilistically associate the appearance detections and the CTF predictions for tracking. Two real world scenarios with 49 ID's were used to evaluate the proposed algorithm. The result show a significant improvement over a baseline tracking algorithm (HSV histogram) and a state-of-the-art online multi person tracking algorithm.
P. Mantini, S. Shah. Camera Placement Optimization Conditioned on Human Behavior and 3D Geometry. pdf
This paper proposes an algorithm to optimize the placement of surveillance cameras in a 3D infrastructure. The key differentiating feature in the algorithm design is the incorporation of human behavior within the infrastructure for optimization. Infrastructures depending on their geometries may exhibit regions with dominant human activity. In the absence of observations, this paper presents a method to predict this human behavior and identify such regions to deploy an effective surveillance scenario. Domain knowledge regarding the infrastructure was used to predict the possible human motion trajectories in the infrastructure. These trajectories were used to identify areas with dominant human activity. Furthermore, a metric that quantifies the position and orientation of a camera based on the observable space, activity in the space, pose of objects of interest within the activity, and their image resolution in camera view was defined for optimization. This method was compared with the state-of-the-art algorithms and the results are shown with respect to amount of observable space, human activity, and face detection rate per camera in a configuration of cameras.
P. Mantini, S. Shah. Enhancing Re-identification Through Contextual Trajectory Forecasting. pdf
Person re-identification (re-ID) is the ability to associate the identity of a person observed at one time and location with the same subject when acquired at a different time and location. Trajectory forecasting is the task of predicting the likely path that a person might take to reach a destination. Contextual trajectory forecasting (CTF) leverages the 3D geometric information of the environment along with observed behavioral norms for human path prediction. Re-ID involves feature matching to find an identity in the database with similar features. The features encompass information regarding appearance of the person like color and texture, or context of the scenario like the time and location of the human subject. CTF provides a future estimate of the likely time and spatial location of previously observed subjects. Embedding this information into traditional re-ID algorithm will significantly boost their performance. In this paper, re-ID is performed across non-overlapping cameras with real world human subjects. CTF is embedded into a re-identification algorithm that uses symmetry driven accumulation of local features (SDALF) to evaluate the performance. Experiments suggest a significant improvement in re-identification by embedding CTF.
P. Mantini, S. Shah. Human Trajectory Forecasting In Indoor Environments Using Geometric Context pdf
A human trajectory is the likely path a human subject would take to get to a destination. Human trajectory forecasting algorithms try to estimate or predict this path. Such algorithms have wide applications in robotics, computer vision and video surveillance. Understanding the human behavior can provide useful information towards the design of these algorithms. Human trajectory forecasting algorithm is an interesting problem because the outcome is influenced by many factors, of which we believe that the geometry of the environment plays a significant role. In addressing this problem, we have built a model to estimate the occupancy behavior of humans based on the geometry and behavioral norms. We also develop a trajectory forecasting algorithm that understands this occupancy and leverages it for trajectory forecasting in previously unseen geometries. We perform experiments to quantify the error between our prediction model and the trajectories obtained from real world human subjects and compare them to state of the art models. Results obtained suggests a significant enhancement in the accuracy of trajectory forecasting by incorporating the occupancy behavior model.