Autonomous racing is an evolving and popular sport among engineering students worldwide who compete against each other to develop the best self-driving race car. In this highly technical competition, lap times and top speeds are the criteria that make the difference between victory and a close second place. The most important technology that determines success in that field is machine vision.
The Application Aristurtle is a Student Team from Thessaloniki, Greece, that designs and manufactures electric race cars to participate in Formula Student competitions across Europe. For the last two years, they have been accompanying the development of a fully autonomous system to be integrated into an electric vehicle in order to replace a human driver's involvement in Formula Student competitions.
A fully autonomous race car without a driver requires advanced and efficient machine vision hardware and software to identify track limits and track conditions. In addition, these vision systems need to precisely detect track markings, which in most cases are represented by plastic cones, to supply reliable information for calculating the optimal acceleration and deceleration as well as defining brake points and the perfect course for the race car. That information allows finer control of both speed and steering actions, contributing to the best possible driving performance, using an independent machine learning system that learns from its inputs. This combination of state-of-the-art technologies leads to increasingly faster lap times, better range figures and a tire management that can far surpass cars handled by a human driver.
The Challenge The most challenging factor in this application is to identify the cones in any weather condition while maintaining a high-resolution image quality to feed the machine learning algorithm. The integrated cameras are responsible for capturing track images that need to be processed by the software. The main goal is to make sure the cones (that ultimately represent the track boundaries), as well as the geolocation of the vehicle on the track and other possible obstacles, are identified with high reliability and at high speeds to be able to control the race car´s acceleration, deceleration and direction. Also, the vision system should be easy to integrate into complex structures, smaller and lighter cameras being preferred in these circumstances.
The Solution Aristurtle opted for two JAI Go Series cameras which were set up in their autonomous race car named Thetis under the main hoop of the vehicle, placed at an angle to each other. Both are used for feeding image data to neural network models for object detection applications. To protect the two cameras from rainy weather, the team manufactured a water-resistant case in order to sustain their continuous use throughout the test runs as well as during the competition itself.
The Benefits Size and weight are critical parameters in the race vehicle design, so the Aristurtle team benefited from the JAI Go series camera´s lightweight and small dimensions when integrating them into their race car. As the cameras are operated at high speeds and trembles, global shutter over rolling shutter cameras was a trivial choice, as they are free of spatial distortions. JAI´s Go Series cameras feature the preferred global shutter. Analog gain control and other features of these cameras guaranteed a maximized image quality within the desired field of view and the appropriate focal length, which also made the Go Series cameras the perfect choice for that application.
The Camera For Aristurtle, the Go Series cameras represent the best solution due to their compactness of only 29 x 29 x 41.5 mm (excluding lens mount), a low weight of only 46 grams, a sturdy housing and its industrial-grade extensive resistance to up to 80G shock and up to 10G vibration, resulting in reliable performance with over 180,000 hours MTBF ratings. Moreover, the team decided to use JAI's cameras because of its enticing capabilities and features, including a USB3 vision interface for plug-and-play convenience.