Towards Practical Evaluation Human Detectors
Where Published: IEEE Transaction on Intelligent Transportation Systems
We introduce a framework for evaluating human detectors that considers the practical application of a detector on a full image using multi-size sliding window scanning. We produce DET (Detection Error Tradeoff) curves relating miss detection rate and false alarm rate computed by deploying the detector on cropped windows as well as whole images, using in the later either image resize or feature resize. Plots for cascade classifiers are generated based on confidence scores instead varying the number of layers. To assess a method's overall performance on a given test, we use the ALMR (Average Log Miss Rate) as an aggregate performance score. To analyze the statistical significance of our results, we conduct 10-fold cross validation experiments. We applied our evaluation framework to two state of the art detectors, on the standard INRIA Person dataset, as well as a local dataset of near infrared images shot from a ground moving vehicle. We applied our evaluation framework to study the differences between the tow detectors on the two datasets with different evaluation methods.
