Adaptive pulsed laser line extraction for terrain reconstruction using a dynamic vision sensor

Abstract

Mobile robots need to know the terrain in which they are moving for path planning and obstacle avoidance. This paper proposes the combination of a bio-inspired, redundancy-suppressing dynamic vision sensor (DVS) with a pulsed line laser to allow fast terrain reconstruction. A stable laser stripe extraction is achieved by exploiting the sensor's ability to capture the temporal dynamics in a scene. An adaptive temporal filter for the sensor output allows a reliable reconstruction of 3D terrain surfaces. Laser stripe extractions up to pulsing frequencies of 500 Hz were achieved using a line laser of 3 mW at a distance of 45 cm using an event-based algorithm that exploits the sparseness of the sensor output. As a proof of concept, unstructured rapid prototype terrain samples have been successfully reconstructed with an accuracy of 2 mm.

Robust 3D Acquisition Using Motion Contrast 3D Scanning

“…The event timing from the DVS is used to determine scan angle, establishing projector-camera correspondence for each pixel. The DVS was used previously for SL scanning by Brandli et al [5] in a pushbroom setup that sweeps an affixed camera-projector module across the scene. This technique is useful for large area terrain mapping, but ineffective for 3D scanning of dynamic scenes.…”

Event-Based Visual Data Sets for Prediction Tasks in Spiking Neural Networks

“…Such devices require low bandwidth and power, making them ideal for robot embodiment. The DVS has been used for classification tasks including recognising hand gestures [4], classifying the DVS version of the MNIST postcode data set [5] and extracting pulsed laser line for terrain reconstruction [6]. DVS data have also been used in tracking tasks, such as tracking people [7], particles [8], optic flow [9], ball and car trajectories [10,11] and LED markers with the camera mounted on a robot [12].…”

Inspired by nature: timescale-free and grid-free event-based computing with spiking neural networks

“…There are a variety of ways to handle time and space in existing event-based sensory data studies (for a detailed review see Section 2.3). Some studies group events into frames or time slices imposing fixed timescales [6][7][8][9][10][11][12][13][14][15]. Other studies process the data stream on an event-by-event basis [16][17][18][19][20][21][22][23][24][25][26][27][28][29], with the timing of an event determining its relevance to an internal model.…”

“…• Hand gesture recognition [6], different data sets with stereo cameras [32] • MNIST post code digit classification [8,29,33] • Classifying feature maps generated by spike input events [41] • Recognising rotating propellers and symbols on flipping cards [24,31,42] • Classifying if a motion performed by a human is a fall [43] • Object (cup, ball and box) recognition in dark environment [44] • Full body gesture recognition with stereo cameras [14] • Moving and non-moving object (balls and books) recognition with stereo cameras embedded with the iCub robot [21] • Recognising patterns of cars passing under a bridge over a freeway [45,46] • Detecting the proximity of hands and touch screens and different types of light source recognition [7] • Recognition of pulsed laser line extraction from terrain construction [11] • Classifying dynamic objects with a moving camera [39] Continued on next page Table 2.1 -continued from previous page Task Event-based visual data studies…”

“…Another study has integrated relevant events into images with non-fixed time intervals [9]. Some studies divided the event stream into time slices so that the events within a time slice were processed as a batch, with their timing information preserved and used for computation [10][11][12][13][14][15]. One study accumulated events into packets based on event-rate [56]; the events within a packet were processed together once the packet reached a certain size.…”