Embedded sensor fusion deep network for road user detection

在过去的十年中，人工智能、传感器技术、低功耗处理单元和连接等领域取得了巨大的进步。因此，自动驾驶汽车成为非常有前途和富有成果的技术，吸引了大量的技术巨头，初创公司以及来自世界各地的研究人员。自动驾驶汽车是一种集成了全自动驾驶系统的汽车，可以在道路上自行驾驶和导航，无需任何人为干预，并可以对环境状况做出响应。感知是任何自动驾驶系统中的关键模块。在本研究计划中，我们解决的问题的检测和定位的道路使用者（车辆，行人，骑自行车的人等）。智能车辆通常配备有三种主要的传感器类型：摄像头、激光雷达（或激光扫描仪）和雷达。所有这些传感器本质上都是非常互补的，一个传感器的优势可以弥补另一个传感器的弱点。然而，在具有挑战性的环境条件下，例如恶劣天气，给定传感器的可操作性可能会受到严重影响，并且检测结果可能变得不可靠。为了解决这个问题，感知系统的性能可以通过组合来自不同模态的信息以克服单个传感器的缺点来显著提高。该研究项目将专注于设计，部署，优化和嵌入卷积神经网络（CNN），用于先进驾驶员辅助系统和自动驾驶汽车背景下的传感器融合。

During the past ten years, tremendous progress has been achieved in the fields of artificial intelligence, sensor technology, low-power processing units and connectivity. As a result, self-driving vehicles emerge as very promising and fruitful technology attracting a large number of technology giants, startups, as well as researches from all aroundthe world. An autonomous vehicle is one that incorporates a fully automated driving system, that can drive and navigate on the road by itself, without requiring any human intervention, and can respond to environmental situations. Perception is a critical module in any automated driving system. In this research project, we address the problem of the detection and localization of the road users (vehicles, pedestrians, cyclists, etc.) surrounding the driving environment of an intelligent vehicle.Intelligent vehicles are generally equipped with three main sensor types: cameras, lidar (or laser scanner) and radar. All these sensors are intrinsically very complementary and the strength of one can compensate for the weakness of the other. However, under challenging environmental conditions, such as adverse weather, the operationality of a given sensor might be severely impacted and the detection results may then become unreliable. To solve this issue, the performance of a perception system can be significantly improved by combining the information coming from different modalities to overcome the shortcomings of individual sensors. This research project will focus on the design, deployment, optimization and embedding of a convolutional neural network (CNN) for sensor fusion in the context of advanced driver assistance systems and autonomous vehicles.