Under unsteady weather conditions (gusty wind and partial cloudiness), the pixel intensities measured by infrared or optical imaging sensors may change considerably within even minutes. This makes a principal obstacle to automated target detection and recognition in real, outdoor settings. Currently existing automated recognition algorithms require strong similarity between the weather conditions of training and recognition. Empirical attempts to normalize image intensities do not lead to reliable detection in practice (e.g. for scenes with a complex relief). Also if the weather is relatively stable (weak wind, rare clouds), as short as 15-20 minutes delay between the training survey and the recognition survey may badly affect target recognition or detection, unless the targets are well separable from background. Thermal IR technologies based on invariants such as emissivity and thermal inertia are expensive and ineffective in making the recognition automated.
Our approach to overcoming the problem is to take advantage of multitemporal prior surveying. It exploits the fact, that any new infrared or optical image of a scene can be accurately predicted based on sufficiently many scene images acquired previously. This removes the above severe constraints to variability of the weather conditions, whereas neither meteorological measurement nor radiometric calibration of the sensor are required. The present paper further generalizes the approach and addresses several points that are important for putting the ideas in practice. Two experimental examples: intruder detection and recognition of a suspicious target illustrate the potential of our method.
This paper presents two approaches to ATR* by trainable algorithms. The first approach assumes that the measurements coming from the objects remain unchanged for the time passed between the stages of learning and recognition. For outdoor scenes such an approach is viable when both learning and recognition can be completed within minutes, which is difficult to achieve in practice. More realistic is to acquire training image data short before surveying the scene of interest. Then computer-intensive or interactive learning algorithms can be applied. We exemplify this approach qualitatively by detecting buildings and asphalt roads in a typical urban scene from AISA hyperspectral sensor data. The second, new approach we derive takes into account the joint changes of all targets and backgrounds under dynamic external factors. This requires multitemporally surveying an area that is specially selected for training an ATR system. Then at the future recognition stage the system can take advantage of the learning results in the real-time mode. Experimental verification of the new approach was performed using a fixed FLIR-type camera that surveyed the site containing more than 50 thermally different objects, whereas learning and recognition were spaced one week apart. The thermal joint prediction model proved working and was applied for detecting and identifying a scene anomaly -- an intruder.