One way to increase the robustness and efficiency of unmanned aerial surveillance platforms is to introduce an autonomous data acquisition capability. In order to mimic a sensor operator´s search pattern, combining wide area search with detailed study of detected regions of interest, the system must be able to produce target indications in real time. Rapid detection algorithms are also useful for cueing image analysts that process large amounts of air reconnaissance imagery, as well as in target seekers. The computational efficiency of a detector is affected by its structural design, the cost of feature extraction, and the discriminative power of the features. The use of a sequence (or cascade) of increasingly complex classifiers has by several authors been suggested as a means to achieve high processing rates at low false alarm and miss rates. The basic principle (applied recursively) is that much of the background can be rejected by a simple classifier before a more complex classifier is applied to the more difficult remaining image regions. The training of each stage of the detector cascade can be integrated with a feature selection process that iteratively adds the most discriminative features from a very large set of features until the detection performance criteria are met. The features used in this study are similar to Haar wavelets and can be computed extremely efficiently. The classifier is trained using a variant of the LogitBoost algorithm that permits different penalties for false alarms and misses. The results obtained are encouraging, and suggest that it is possible to achieve high-processing rates at very low false alarm and miss rates when searching for military ground vehicles.