A car that is designed to detect crosswalks
Many pedestrians pass away because they are hit by car, especially on crosswalks. Drivers that don't see the crosswalk, pedestrians that don't check the street, pass the street on red light, or cross the street illegally, drivers that drive with high speed, or even the absence of road signs are just a number of reasons why accidents happen. Even though traffic rules and punishments exist, The Governors Highway Safety Administration calculates that there were 6590 pedestrian fatalities in 2019 after making a preliminary analysis of the available data. That would mark a 60 percent increase in fatalities since 2009.
How can the problem be solved?
SMV is a car that wants to combat this issue. It can do this by automatization and robotization of the car board. By implemmenting different technologies to detect both crosswalks and pedestrians, it can surpass the human limits.
How is the car equipped?
The car is equipped with 3 frontal cameras, a radar, movement senzors, AI and of course, a software that integrates all technologies in one place: the car's board. The AI will recognize the crosswalks and the pedestrians, having the information provided by the other components.
Artificial Intelligence
I believe that an approach based on advanced AI for vision, supported by efficient use of inference hardware is a good way to achieve a general solution for detecting crosswalks and pedestrians.
Apply research to train deep neural networks on problems like perception. SMV cameras analyze raw images to perform semantic segmentation, object detection and monocular depth estimation. The car's network take video from all cameras to output the road layout, static infrastructure and 3D objects. In this way, the car is able to detect crosswalks, road signs and pedestrians.
Algorithms
In order to train the neural networks to predict such representations, algorithmically create accurate and large-scale ground truth data by combining information from the car's sensors across space and time. By building a robust planning and decision-making system that operates in real-world we can deal with unexpected situations.
Code
Build the Autopilot software foundation up from the lowest levels of the stack, tightly integrating with hardware. Write fast, memory-efficient code to capture high-frequency, high-volume data from sensors, and to share it with multiple consumer processes— without impacting central memory access latency or starving critical functional code from CPU cycles.
Evaluation infrastructure
Write code simulating our real-world environment, producing highly realistic graphics and other sensor data that feed our Autopilot software for live debugging or automated testing.
3 Cameras
SMV has 3 cameras faced front, in order to capture the world in 3 different styles: a camera that is wide and has a distance of 50m, a main camera with a distance of 150m and a narrow one with a distance of 250m.
Radar
The car also has a forward-facing radar that provides additional data about the world on a redundant wavelength that is able to see through heavy rain, fog, dust and even the car ahead.
Movement sensors
A magnetometer works by using a passive sensing technology to detect large ferrous objects (for example, a truck, automobile, or rail car) by measuring the change in the ambient magnetic field. When a vehicle alters that magnetic field, the sensor detects those changes. The range of the magnetometer will depend on the target.
