Building management system using object detection and tracking in a large space with a low resolution sensor
A method of operating an object detection and tracking system includes the step of estimating ( 202 ) a current background of a current frame of sensor data generated by a sensor based on a previous frame of sensor data by a computer-based processor. The method further includes estimating ( 204 ) a foreground of the current frame of sensor data by comparing the current frame of sensor data to the current background, and detecting ( 212 ) an object using a sensor-specific object model.
1. A method of operating an object detection and tracking system comprising:
estimating a current background of a current frame of sensor data generated by a sensor and based on a previous frame of sensor data by a computer-based processor;
estimating a foreground of the current frame of sensor data by comparing the current frame of sensor data to the current background; and detecting an object using a sensor-specific object model, wherein the sensor is an absolute intensity sensor utilizing a chopper; and
tracking the object via a Bayesian Estimator, and wherein the sensor-specific object model is a chopped-data object model, wherein the chopped-data object model is learned by discriminative dictionary learning.
2. The method set forth in claim 1 , wherein the chopped-data object model is a Gaussian Mixture object model.
3. The method set forth in claim 1 , wherein the chopped-data object model is parameterized at least in-part by perspective data.
4. The method set forth in claim 1 , wherein the Bayesian Estimator is a Kalman Filter.
5. The method set forth in claim 1 , wherein the Bayesian Estimator is a Particle Filter.