Biologically Inspired Vision for Indoor Robot Navigation

Abstract

Ultrasonic, infrared, laser and other sensors are being applied in robotics. Although combinations of these have allowed robots to navigate, they are only suited for specific scenarios, depending on their limitations. Recent advances in computer vision are turning cameras into useful low-cost sensors that can operate in most types of environments. Cameras enable robots to detect obstacles, recognize objects, obtain visual odometry, detect and recognize people and gestures, among other possibilities. In this paper we present a completely biologically inspired vision system for robot navigation. It comprises stereo vision for obstacle detection, and object recognition for landmark-based navigation. We employ a novel keypoint descriptor which codes responses of cortical complex cells. We also present a biologically inspired saliency component, based on disparity and colour.

Biologically Inspired Vision for Human-Robot Interaction

“…Finally, we apply a threshold vector, also previously trained on the Notredame dataset, to set each of the 128 bits to 1 or 0. The resulting descriptor is a huge improvement of the previous one [17]. It is comparable to the SIFT-based LDAHash descriptor in terms of performance when tested on the Yosemite dataset [19].…”

“…Also, from a computational point of view, a binary descriptor is much faster to compute and to match than a floating-point one. This method is an improvement of the previous method [17]. We start by applying maximum pooling in a circular region around each keypoint, followed by zero-mean normalization and extraction of the maximum cell responses in 8 filter orientations.…”

“…It can also be useful for computing visual saliency. The algorithm we used to generate the disparity maps is the same as previously used in [17]: (a) resize the left and right images to a small size (160 × 120); (b) extract complex cell responses on circles around each pixel; (c) compare each pixel P in the left image to the next K pixels in the right image on the same line and starting from the same position P as in the left image (we used K = 35); (d) use the Hamming distance to find the best-matching pixels; and (e) apply median filtering (5 × 5 kernel) to reduce noise due to wrong matches. The computed disparity maps are then thresholded to find nearby obstacles (t d = 70).…”

“…The generated saliency maps are also useful to segregate hands from the background, to reduce clutter and to improve gesture recognition rates. The visual saliency algorithm we developed is an improvement of the one described in [17] and combines three different features: color, disparity and motion. The three features are processed separately and then merged into a single saliency map with equal weights.…”