Sumio Yokomitsu

Shrink boost for selecting multi-LBP histogram features in object detection

Feature selection from sparse and high dimension features using conventional greedy based boosting gives classifiers of poor generalization. We propose a novel “shrink boost” method to address this problem. It solves a sparse regularization problem with two iterative steps. First, a “boosting” step uses weighted training samples to learn a full high dimensional classifier on all features. This avoids over fitting to few features and improves generalization. Next, a “shrinkage” step shrinks least discriminative classifier dimension to zero to remove the redundant features. In our object detection system, we use “shrink boost” to select sparse features from histograms of local binary pattern (LBP) of multiple quantization and image channels to learn classifier of additive lookup tables (LUT). Our evaluation shows that our classifier has much better generalization than those from greedy based boosting and those from SVM methods, even under limited number of train samples. On public dataset of human detection and pedestrian detection, we achieve better performance than state of the arts. On our more challenging dataset of bird detection, we show promising results.

A layered-based fusion-based approach to detect and track the movements of pedestrians through partially occluded situations

To obtain perception abilities, conventional methods independently detect static and dynamic obstacles, and estimate their related information, which is not quite reliable and computationally heavy. We propose a fusion-based and layered-based approach to systematically detect dynamic obstacles and obtain their location and timing information. The layered-based concept helps us to first search pedestrians in horizontal dimension based on transitional peaks in the defined projection-curves, and then search in vertical dimension. Converting a typical 2D search problem into two 1D search problems significantly decreases the computational load. The fusion-based obstacle detection fuses the information from initial segmentation and dynamic tracking model to avoid complicated tracking schemes. The methodologies take advantage of connection between different information, and increase the accuracy and reliability of obstacle segmentation and tracking. The search mechanism works for both visible and infrared sequences, and is specifically effective to track the movements of pedestrians in complicated environments such as human intersecting and conclusion, thus improving environment understanding abilities and driving safety.