Search

VIDET is a research project active at the University of Bologna and aimed at the development of a mobility aid for the visually impaired. VIDET's approach consists in the conversion of depth data gathered through a stereo-vision system into a 3D model perceivable by the user by means of a wire-actuated haptic interface. In this paper we describe VIDET's PC-based, real-time, stereo-vision system. As for system's description, we outline the structure of the stereo-matching algorithm and address in more detail the optimization strategies that lead us to a fast PC-based implementation. These involve massive reduction of redundant calculations and use of the parallel multimedia instructions available in current general-purpose microprocessors. We provide experimental results showing that the system is capable of recovering correctly the 3D structure of the observed scene and allows for prompt perception of the depth changes generated by moving objects. We also report execution time measurements and compare our stereo system with the PC-based systems from SRI and Point Grey Research. © 2002 Elsevier Science Ltd. All rights reserved.

Local stereo matching algorithms based on the adapting-weights strategy achieve accuracy similar to global approaches. One of the major problems of these local algorithms is that they are computationally expensive. How- ever, algorithms with reduced computational complexity in- spired by the adapting-weights strategy have been recently proposed. In particular, the Fast Bilateral Stereo (FBS) framework allows to obtain, with a significantly reduced computational burden, results comparable to top-performing local approaches based on adapting-weights. In this paper we propose a novel framework that has two advantages: en- ables a further speedup of this type of algorithms along with a slight accuracy improvement. We prove the effectiveness of our proposal in combination with the FBS approach.

This paper is aimed at evaluating the performances of different measures which have been proposed in literature for robust matching. In particular, classical matching metrics typically employed for this task are considered together with specific approaches aiming at achieving robustness. The main aspects assessed by the proposed evaluation are robustness with respect to photometric distortions, noise and occluded patterns. Specific datasets have been used for testing, which provide a very challenging framework for what concerns the considered disturbance factors and can also serve as testbed for evaluation of future robust visual correspondence measures.

This paper deals with the scarcity of data for training optical flow networks, highlighting the limitations of existing sources such as labeled synthetic datasets or unlabeled real videos. Specifically, we introduce a framework to generate accurate ground-truth optical flow annotations quickly and in large amounts from any readily available single real picture. Given an image, we use an off-the-shelf monocular depth estimation network to build a plausible point cloud for the observed scene. Then, we virtually move the camera in the reconstructed environment with known motion vectors and rotation angles, allowing us to synthesize both a novel view and the corresponding optical flow field connecting each pixel in the input image to the one in the new frame. When trained with our data, state-of-the-art optical flow networks achieve superior generalization to unseen real data compared to the same models trained either on annotated synthetic datasets or unlabeled videos, and better specialization if combined with synthetic images., Comment: CVPR 2021. Project page with supplementary and code: https://mattpoggi.github.io/projects/cvpr2021aleotti/

Depth estimation is crucial in several computer vision applications, and a recent trend aims at inferring such a cue from a single camera through computationally demanding CNNs — precluding their practical deployment in several application contexts characterized by low-power constraints. Purposely, we develop a tiny network tailored to microcontrollers, processing low-resolution images to obtain a coarse depth map of the observed scene. Our solution enables depth perception with minimal power requirements (a few hundreds of mW), accurately enough to pave the way to several high-level applications at-the-edge.

Among the most straining trends that farmers have to face there are: on one side, to guarantee welfare and adequate life conditions for animals and to reduce the environmental footprint, on the other side, to develop new strategies to improve farm management reducing costs. The current conditions and the expected developments of the dairy sector highlight a strong need for more efficient and sustainable farming systems. Studying heat stress, herd management and housing and animals’ productive and reproductive performances is fundamental for the economic and environmental sustainability of the dairy chain. New and effective tools to cope with these challenges have been provided by Precision Livestock Farming (PLF), which is nowadays increasingly applied and makes possible to control quali-quantitative parameters related to production, health, behaviour, and real-time locomotion per animal. The research key challenge is to turn these data into knowledge to provide real-time support in farming optimisation. This research focuses specifically on different systems to collect, process and derive useful information from data on animal welfare and productivity. A multi-disciplinary approach has been adopted to generate a decision support system for farmers.

This work deals with the implementation of energy-efficient monocular depth estimation using a low-cost CPU for low-power embedded systems. It first describes the PyD-Net depth estimation network, which consists of a lightweight CNN able to approach state-of-the-art accuracy with ultra-low resource usage. Then it proposes an accuracy-driven complexity reduction strategy based on a hardware-friendly fixed-point quantization. Finally, it introduces the low-level optimization enabling effective use of integer neural kernels. The objective is threefold: (i) prove the efficiency of the new quantization flow on a depth estimation network, that is, the capability to retaining the accuracy reached by floating-point arithmetic using 16- and 8-bit integers, (ii) demonstrate the portability of the quantized model into a general-purpose 32-bit RISC architecture of the ARM Cortex family, (iii) quantify the accuracy-energy tradeoff of unsupervised monocular estimation to establish its use in the embedded domain. The experiments have been run on a Raspberry PI board powered by a Broadcom BCM2837 chipset. A parametric analysis conducted over the KITTI date-set shows marginal accuracy loss with 16-bit (8-bit) integers and energy savings up to 6.55× (9.23×) w.r.t. floating-point. Compared to high-end CPU and GPU the proposed solution improves scalability.

As reported in the recent image velocimetry literature, tracking the motion of sparse feature points floating on the river surface as done by the Optical Tracking Velocimetry (OTV) algorithm is a promising strategy to address surface flow monitoring. Moreover, the lightweight nature of OTV coupled with computational optimizations makes it suited even for its deployment in situ to perform measurements at the edge with cheap embedded devices without the need to perform offload processing. Despite these notable achievements, the actual practical deployment of OTV in remote environments would require cheap and self-powered systems enabling continuous measurements without the need for cumbersome and expensive infrastructures rarely found in situ. Purposely, in this paper, we propose an additional simplification to the OTV algorithm to reduce even further its computational requirements, and we analyze self-powered off-the-shelf setups for in situ deployment. We assess the performance of such set-ups from different perspectives to determine the optimal solution to design a cost-effective self-powered measurement node.

Confidence measures for stereo earned increasing popularity in most recent works concerning stereo, being effectively deployed to improve its accuracy. While most measures are obtained by processing cues from the cost volume, top-performing ones usually leverage on random-forests or CNNs to predict match reliability. Therefore, a proper amount of labeled data is required to effectively train such confidence measures. Being such ground-truth labels not always available in practical applications, in this paper we propose a methodology suited for training confidence measures in a self-supervised manner. Leveraging on a pool of properly selected conventional measures, we automatically detect a subset of very reliable pixels as well as a subset of erroneous samples from the output of a stereo algorithm. This strategy provides labels for training confidence measures based on machine-learning technique without ground-truth labels. Compared to state-of-the-art, our method is neither constrained to image sequences nor to image content. Experimental results on three challenging datasets with three stereo algorithms and three state-of-the-art confidence measures based on machine-learning techniques confirm the effectiveness of our proposal for self-supervised training.

This paper proposes an effective and wearable mobility aid aimed at improving the quality of life of people suffering for visual disabilities by enabling autonomous and safe navigation in unknown environments. Our system relies on dense and accurate depth maps, provided in real-time by a compact stereo vision system mapped into an FPGA, in order to detect obstacles in front of the user and to provide accordingly vibration feedbacks as well as audio information by means of a bone-conductive speakers. Compared to most approaches with similar purposes, even in the current prototype arrangement deployed for testing, our system is extremely compact, lightweight and energy efficient thus enabling hours of safe and autonomous navigation with standard batteries avoiding the need to carry cumbersome devices. Moreover, by conceiving the 3D sensing device as a replacement of standard glasses typically worn by visually impaired people and by using intuitive feedbacks provided by means of lightweight actuators, our system provides an ergonomic and comfortable user interface with a fast learning curve for its effective deployment. This fact has been extensively verified on the field by means of an experimental evaluation, in indoor as well as in outdoor environments, with different users simulating visual impairment including a blind person.

Thank to the widespread diffusion of RGBD sensing devices based on active technologies, in recent years, many research and industrial applications have taken advantage of reliable cues computed from dense depth data. However, although these sensors can be very effective in many circumstances, they are not always well suited for outdoor environments and can also interfere with each other when sensing the same area. On the other hand, traditional systems based on passive stereo vision technology, due to their computational/energy requirements, reliability, size, cost etc, have been, so far, mostly confined to research projects. Nevertheless, recent advances in computation architectures and algorithms enable to overcome most of these issues and, in this paper, we describe the architecture and the processing pipeline of an effective RGBD sensor based on stereo vision suited for real time applications. This sensor allows us to infer, in indoor and outdoor environments, dense and accurate depth maps computed according to state-of-art algorithms and with minimal energy requirements that fit with typical constraints of smart camera systems.

People tracking is a crucial component of many intelligent video surveillance systems and recent developments in embedded computing architectures and algorithms allow us to design compact, lightweight and energy efficient systems aimed at tackling this problem. In particular, the advent of cheap RGBD sensing devices enables to exploit depth information as additional cue. In this paper we propose a 3D tracking system aimed to become the basic node of a distributed system for business analytics applications. In the envisioned distributed system each node would consist of a custom stereo camera with on-board FPGA processing coupled with a compact CPU based board. In the basic node proposed in this paper, aimed at raw people tracking within the sensed area of a single device, the custom stereo camera delivers, in real time and with minimal energy requirements, accurate dense depth maps according to state-of-the-art computer vision algorithms. Then, the CPU based system, by processing this information enables reliable 3D people tracking. In our system, deploying the FPGA front-end, the main constraint for real time 3D tracking is concerned with the computing requirement of the CPU based board and, in this paper, we propose a fast and effective node for 3D people tracking algorithm suited for implementation on embedded devices.