No abstract

Object detection is an essential technology for surveillance systems, particularly in areas with a high risk of accidents such as railway level crossings. To prevent future collisions, the system must detect and track any object that passes through the monitored area with high accuracy, and this process must be performed fulfilling real-time specifications. In this work, an edge IoT HW platform implementation capable of detecting and tracking objects in a railway level crossing scenario is proposed. The response of the system has to be calculated and sent from the proposed IoT platform to the train, so as to trigger a warning action to avoid a possible collision. The system uses a low-resolution 3D 16-channel LIDAR as a sensor that provides an accurate point cloud map with a large amount of data. The element used to process the information is a custom embedded edge platform with low computing resources and low-power consumption. This processing element is located as close as possible to the sensor, where data is generated to improve latency, privacy, and avoid bandwidth limitations, compared to performing processing in the cloud. Additionally, lightweight object detection and tracking algorithm is proposed in this work to process a large amount of information provided by the LIDAR, allowing to reach real-time specifications. The proposed method is validated quantitatively by carrying out implementation on a car road, emulating a railway level crossing.

Nowadays, with the huge advance of sensor technology and the increase of the amount of data generated by them, techniques have to be developed to be able to process all this amount of information in real-time applications on edge devices, close to where data is being generated. If all that information has to be sent to the cloud to be processed, it has certain disadvantages in terms of latency, bandwidth, privacy and reliability, compared to locally processing it on the edge. In this paper, the implementation of deep learning algorithms in low power and limited resources devices in an Internet of Things scenario is studied. In order to work in real-time applications, the influence of different low power consumption deep learning hardware accelerators is studied. Finally, a practical case for smart farming is shown with comparative results in terms of power consumption and performance when running the same artificial vision algorithm on different devices.

The current trend of shifting computing from the cloud to the edge of the Internet of Things is influencing deep learning applications. Moving intelligence closer to the point of need entails advantages in terms of performance, power consumption, security, and privacy. The problem arises with data sources that generate a massive amount of information, making data processing challenging for edge devices. This is the case of point clouds generated by LIDAR sensors. Implementations at the edge become even more challenging when heavy processing algorithms such as deep neural networks are selected. However, deep neural networks are the stateof-the-art solution to carry out object classification tasks as they provide the best results in terms of accuracy when working with high data volumes. This work demonstrates that the processing of point cloud-based sensors using deep neural networks at the edge is becoming feasible with the emergence of new devices with high computing capacity combined with reduced power consumption. In this regard, a characterization of first-in-class deep learning classification algorithms working with point cloud data as inputs and running over different state-of-the-art edge processing architectures is provided. A broad range of devices, including CPUs, GPU-based, SoC FPGA-based, and deep learning neural accelerators, have been evaluated in terms of inference time, classification accuracy, and power consumption. As a result, it demonstrates that neural accelerators with integrated host CPUs represent the best trade-off between power consumption and performance, making them a perfect solution for IoT applications at the edge level.

LiDAR sensors are increasing in popularity due to the advantages they provide over 2D sensors in IoT object detection and classification applications, because of their ability to provide very precise distances to objects. Deep learning algorithms need a huge amount of data during training to obtain high accuracy results. When using 2D images a vast quantity of datasets are publicly available, but this is not the case for LiDAR point clouds. Each LiDAR model generates a point cloud with unique properties, which causes the datasets not to be compatible between different LiDAR models. As a result, when using deep learning with LiDARs it is necessary to generate the datasets manually. For this purpose, the data must be captured and then labeled one by one, which is a very time and cost consuming process. To overcome this issue and to reduce the development time when using LiDAR sensors with deep learning algorithms, a methodology is proposed in this paper to automatically generate point cloud datasets using a 3D simulator for autonomous cars. In this regard, a dataset can be generated for any LiDAR model by adding the specific LiDAR parameters to the simulator. Besides, custom scenarios can be designed and generated, based on the final deployment location, to provide a simulated solution very close to the final implementation. With the proposed methodology, a simulation can be performed to select the LiDAR that best fits certain application requirements, in contrast to the traditional approach where the LiDAR must first be purchased.