1354 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 45, NO. 7, JULY 2010 A 134-Pixel CMOS Sensor for Combined Time-of-Flight and Optical Triangulation 3-D Imaging Oreste Sgrott, Daniel Mosconi, Matteo Perenzoni, Member, IEEE, Gianmaria Pedretti, Lorenzo Gonzo, Member, IEEE, and David Stoppa, Member, IEEE Abstract—This paper describes the design and characterization of a 134-pixel linear array sensor for three-dimensional measure- ments based on both multiple-pulse indirect-time-of-flight (ITOF) and optical triangulation (OT) techniques. In OT mode, a winner- take-all (WTA) stage allows for a fast localization of the spot posi- tion along the pixel array so that only useful pixels are selectively read out, for a maximum operation speed of 131 kVoxel/s. Distance measurements in OT mode over the range 0.4 m–1.0 m are obtained with a best precision of 0.004%–0.21%, while ITOF operation allows mapping the range 0.8–3 m at 125 voxel/s with a relative precision of 1.7%–3.8%. Background rejection up to 10 klux has also been demonstrated without the need of any optical filters. Index Terms—CMOS active pixel, image sensor, range finding, three-dimensional image sensor, optical triangulation, time-of-flight. I. INTRODUCTION B ECAUSE we are living in a three-dimensional world, the information provided by off-the-shelf digital cameras is often not sufficient to build the sophisticated models required by systems capable of analyzing and interpreting the environment. A three-dimensional (3-D) vision tool could offer amazing pos- sibilities of improvement in many areas like automotive appli- cations, security and surveillance, cultural heritage preserva- tion, hazardous sites inspection, industrial control, cinema in- dustry, virtual reality, etc., because (i) it significantly increases the robustness of object classification, (ii) it allows for using a lower resolution sensor type, and (iii) it avoids time-consuming post-processing steps. For this reason, a large number of 3-D acquisition systems have been developed in the last years. Optical distance mea- suring techniques can be classified mainly into three categories: interferometry, triangulation and time-of-flight (TOF) [2]. Non- optical techniques (e.g., radar, ultrasonic) are not considered here because of their extremely poor performance in terms of lateral resolution. Interferometry allows for very high precision Manuscript received November 24, 2009; revised February 15, 2010; ac- cepted March 15, 2010. Date of current version June 25, 2010. This paper was approved by Guest Editor Kofi Makinwa. This work was supported by the Au- tonomous Province of Trento, Italy, within the project 3D-ARCH. The authors are with Fondazione Bruno Kessler (FBK), I-38100 Povo, Trento, Italy (e-mail: stoppa@fbk.eu). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JSSC.2010.2048076 in the distance estimation, basically related to the wavelength of the coherent light source used to actively illuminate the scene [3]. However, interferometric systems require bulky and precise optical setup and the depth range is limited. Triangulation tech- niques (e.g., stereo vision, laser triangulation and fringe projec- tion) offer excellent performance in terms of accuracy, but the maximum depth range is limited since it is determined by the triangulation baseline length [4]. TOF techniques are based on the measurement (either direct or indirect) of the time needed for an optical signal to travel from a source to a target and back to a sensor. TOF techniques provide the best performance in terms of acquisition speed, distance range and overall cost of the system when scannerless architectures are used, but exhibit lower precision performance. Many new developments in the field of CMOS image sensors dedicated to 3-D imaging have been reported, mostly focused on TOF principle. The most mature solution exploits photo-de- modulators, in which the photogenerated charge is “mixed” on two or more collection electrodes thus achieving an intrinsic de- modulation effect [5]–[10]. Pioneering works in this field were based on CCD technology [5], [6]. The same basic concept has been further developed leading to the first 2-D pixel array re- ported in [7], and yielding a very good overall performance, both in terms of accuracy (better than 5 cm) and of maximum mea- surable range (7.5 m). A refined version of this chip represents one of the first examples of 3-D imager commercially available [8] where a 4 mm @ 2 m (0.2%) precision is attained at 30 fps. Also based on photo-demodulation is the sensor reported in [9], which is fabricated with a modified CMOS process to improve charge mixing efficiency. It provides very good results in terms of pixel size (the pitch is only 15 m) and array size (336 252); the best resolution achieved is 2.35 cm @ 1.6 m (1.5%) at 30 fps. Charge draining structures are included in the pixel to reduce background light influence. Another promising solution exploits single-photon avalanche diodes (SPADs) integrated within CMOS technologies. Prob- lems common to SPAD-based imagers are the relatively large achievable pitch along with a small fill factor, and the high sensi- tivity to ambient light. In [11] a 32 32-pixel array is reported, having a 58 m pitch. Distance can be measured on a range up to 3 m with a precision of 4.6 mm @ 3 m (0.15%) at 0.49 fps. The main limit is that each pixel must be addressed serially, thus delaying the measurement by a few seconds or minutes, along with the need for an external time-to-digital converter (TDC). The solution proposed in [12] solves these problems 0018-9200/$26.00 © 2010 IEEE