Mehmet Kemal Kocamaz

Appearance contrast for fast, robust trail-following

We describe a framework for finding and tracking “trails” for autonomous outdoor robot navigation. Through a combination of visual cues and ladar-derived structural information, the algorithm is able to follow paths which pass through multiple zones of terrain smoothness, border vegetation, tread material, and illumination conditions. Our shape-based visual trail tracker assumes that the approaching trail region is approximately triangular under perspective. It generates region hypotheses from a learned distribution of expected trail width and curvature variation, and scores them using a robust measure of color and brightness contrast with flanking regions. The structural component analogously rewards hypotheses which correspond to empty or low-density regions in a groundstrike-filtered ladar obstacle map. Our systems performance is analyzed on several long sequences with diverse appearance and structural characteristics. Ground-truth segmentations are used to quantify performance where available, and several alternative algorithms are compared on the same data.

Trail following with omnidirectional vision

We describe a system which follows “trails” for autonomous outdoor robot navigation. Through a combination of visual cues provided by stereo omnidirectional color cameras and ladar-based structural information, the algorithm is able to detect and track rough paths despite widely varying tread material, border vegetation, and illumination conditions. The approaching trail region is simply modeled as a circular arc of constant width. Using an adaptive measure of color and brightness contrast between a hypothetical region and flanking areas, the tracker performs a robust randomized search for the most likely trail region and robot pose relative to it with no a priori appearance model. Stereo visual odometry improves tracker dynamics on uneven terrain and permits local obstacle map maintenance. A motion planner is also described which takes the trail shape estimate and local map to plan smooth trajectories around in-trail and near-trail hazards. Our systems performance is analyzed on several long sequences with diverse appearance and structural characteristics using ground-truth segmentations.

Integrating stereo structure for omnidirectional trail following

We describe a system which follows “trails” for autonomous outdoor robot navigation. Through a combination of appearance and structural cues derived from stereo omnidirectional color cameras, the algorithm is able to detect and track rough paths despite widely varying tread material, border vegetation, and illumination conditions. The approaching trail region is modeled as a circular arc segment of constant width. Using likelihood formulations which measure color, brightness, and/or height contrast between a hypothetical region and flanking areas, the tracker performs a robust randomized search for the most likely trail region and robot pose relative to it with no a priori appearance model. The addition of the structural information, which is derived from a semi-global dense stereo algorithm with ground-plane fitting, is shown to improve trail segmentation accuracy and provide an additional layer of safety beyond solely ladar-based obstacle avoidance. Our systems ability to follow a variety of trails is demonstrated through live runs as well as analysis of offline runs on several long sequences with diverse appearance and structural characteristics using ground-truth segmentations.

Vision-based counting of pedestrians and cyclists

This paper describes a vision-based cyclist and pedestrian counting method. It presents a data collection prototype system, as well as pedestrian and cyclist detection, tracking, and counting methodology. The prototype was used to collect approximately 50 hours of data which have been used for training and testing. Counting is done using a cascaded classifier. The first stage of the cascade detects the pedestrians or cyclists, whereas the second stage discriminates between these two classes. The system is based on a state-of-the-art pedestrian detector from the literature, which was augmented to explore the geometry and constraints of the target application. Namely, foreground detection, geometry prior information, and temporal moving direction (optical flow) are used as inputs to a multi-cue clustering algorithm. In this way, false alarms of the detector are reduced and better fitted detection windows are obtained. The presented project was the result of a partnership with the City of Pittsburgh with the objective of providing actionable data for government officials and advocates that promote bicycling and walking.

Approaches for automatic low-dimensional human shape refinement with priors or generic cues using RGB-D data

Some human detection or tracking algorithms output a low-dimensional representation of the human body, such as a bounding box. Even though this representation is enough for some tasks, a more accurate and detailed point-wise representation of the human body is more useful for pose estimation and action recognition. The refinement process can produce a point-wise mask of the human body from its low-dimensional representation. In this paper, we tackle the problem of refining low-dimensional human shapes using RGB-D data with a novel and accurate method for this purpose. This algorithm combines low-level cues such as shape and color, and high level observations such as the estimated ground plane, in a multi-layer graph cut framework. In our algorithm, shape prior information is learned by training a classifier. Unlike some existing work, our method does not utilize or carry features from the internal steps of the methods which provide the bounding box, so our method can work on the outputs of any similar shape providers. Extensive experiments demonstrate that the proposed technique significantly outperforms other suitable methods. Moreover, a previously published refinement method is extended by incorporating more generic cues to serve this purpose. A novel and accurate method to refine low dimensional human shape using RGB-D data is proposed.Uses of multiple modalities do not carry any features from the shape provider.Combines low and high level observations jointly in multi-layer graph structureExtensive experiments showed that it outperforms compared suitable algorithms.Also, an existing method is extended by fusing more generic cues for this purpose.

Deformable object shape refinement and tracking using graph cuts and support vector machines

This paper describes several approaches to the problem of obtaining a refined segmentation of an object given a coarse initial segmentation of it. One line of investigation modifies the standard graph cut method by incorporating color and shape distance terms, adaptively weighted at run time to try to favor the most informative cue given visual conditions. We also discuss a machine learning approach based on support vector machines which uses color and spatial features as well. Furthermore, we extend these single-frame refinement methods to serve as the basis of trackers which work for a variety of object types with complex, deformable shapes. Comparative results are presented for several diverse datasets including objects such as trail regions used for robot navigation, hands, and faces.

Automatic Refinement of Foreground Regions for Robot Trail Following

Continuous trails are extended regions along the ground such as roads, hiking paths, rivers, and pipelines which can be navigationally useful for ground-based or aerial robots. Finding trails in an image and determining possible obstacles on them are important tasks for robot navigation systems. Assuming that a rough initial segmentation or outline of the region of interest is available, our goal is to refine the initial guess to obtain a more accurate and detail representation of the true trail borders. In this paper, we compare the suitability of several previously published segmentation algorithms both in terms of agreement with ground truth and speed on a range of trail images with diverse appearance characteristics. These algorithms include generic graph cut, a shape-based version of graph cut which employs a distance penalty, Grab Cut, and an iterative superpixel grouping method.

Unconstrained 1D range and 2D image based human detection

An accurate and computationally very fast multimodal human detector is presented. This 1D+2D detector fuses 1D range scan and 2D image information via an effective geometric descriptor and a silhouette based visual representation within a radial basis function kernel support vector machine learning framework. Unlike the existing approaches, the proposed 1D+2D detector does not make any restrictive assumptions on the range scan positions, thus it is applicable to a wide range of real-life detection tasks. To analyze the discriminative power of the geometric descriptor, a range scan only version, 1D+, is also evaluated. Extensive experiments demonstrate that the 1D+2D detector works robustly under challenging imaging conditions and achieves several orders of magnitude performance improvement while reducing the computational load drastically. In addition, a new multi-modal (LIDAR, depth image, optical image) dataset, DontHitMe, is introduced. This dataset contains 40,000 registered frames and 3,600 manually annotated human objects. It depicts challenging illumination conditions in indoors and outdoors environments and is publicly available to our community.