SafeT-Rex

This is our autonomous car project at AXA Switzerland Hackathon in Sept 2018. 9 crazy guys (2 AA Sure guys, a Crashtest prevention guy, a 3D-printer guy, a strategy guy, an architect, a bee keeper, a hardware+software tinker and a junior :-) will try to build their own autonomous RC car during the 2 days hackathon.

We have lots of experience in totally 31 Hackathons :-)

context: we are in 2030 and autonomous vehicles are reality. We as AXA-hackers would like to show today how this is working. Important for 2030's autonomous cars are the the newest prevention methods to reduce crashs. In 2030 we are not sure what animals still exist and which are new. Be prepared - Grrrrr.

We built it

2019-09-27 14:31 - here is the first successful trial with lane detection - not perfect yet but a start.

lane detection

the hardest thing is the lane detection algorith. checkout our detailed readme

Features

• turn an existing RC car into a software controllable car using Arduino, raspberry pi3, picam
• use line or lane detection to control the car and drive along a short track
• use object detection using any image / feature recognition algorithm (YOLO, OpenCV haar classifier, Tensorflow, etc)
• use distance control mechanism (image and ultra-sonic) to calculate the distance
• 3D-print own funny signs to control the cars speed
• build our own mobile track to train the car and to demo it at the fair / ZERO Gravity

Demo

• show a live demo in 1st person view on the cinema stage :-) No slides.
• some other things .... pssst. surprise

Links to technologies to be used / shared

• install python. tensorflow. jupyter notebooks !(video: https://youtu.be/RUs_qR1hKds)

• build an image classifier with tensorflow in 10‘ !(video: https://youtu.be/QfNvhPx5Px8)

• pi tensorflow object detection !(video: https://github.com/EdjeElectronics/TensorFlow-Object-Detection-on-the-Raspberry-Pi/blob/master/README.md)

• image prediction history snd yolo explained !(video: https://youtu.be/4eIBisqx9_g)

• https://github.com/ramitix/self-driving-car/blob/master/README.md - full self driving demos step by step

• https://pjreddie.com/darknet/yolo/ (video: https://youtu.be/Cgxsv1riJhI) - You only look once - cool and fast

• https://youtu.be/Rgpfk6eYxJA (How To Train an Object Detection Classifier Using TensorFlow 1.5)

• https://www.pyimagesearch.com/2018/04/09/how-to-quickly-build-a-deep-learning-image-dataset/ - how to build image data set fast using bing search

Open CV

• debugging: https://www.pyimagesearch.com/2014/12/22/visual-logging-new-favorite-tool-debugging-opencv-python-apps/
• extract text with opencv https://www.pyimagesearch.com/2018/09/17/opencv-ocr-and-text-recognition-with-tesseract/
• Training your own detector https://youtu.be/WEzm7L5zoZE

Computer

• we all use our private laptops (windows or mac)
• wifi router: Lolo

Preparation by team

• Auto: Bernhard, Stefan, Lolo
• Pi3: Cris, Baxter, Lolo
• PiCam: Lolo, Baxter
• Linefollower: Bernhard
• Ultraschall, Lötkolben, Zinn, Krimpen: Salvador
• Linoleum oder anderes, dunkel: Cris
• Klebeband, PanzerTape: schwarz, gelb: Stefan
• Heissleim: Lolo
• Akku: Stefan
• Karton: Lolo
• Powerbank, kleine für Auto / Raspi: alle
• Kabelbinder: Stefan
• Kupferdraht: Stefan
• 3D-Drucker: Salvador
• 3 Schilder drucken basierend auf Auto Grösse: Salvador
• iPhone6 alt: Lolo
• Flipchart: AXA
• AppleTV: Lolo

Our AXA hack solution

here is the first successful trial with lane detection - not perfect yet but a start.

highlevel architecture

tbd

Hardware

we are using our own car from the Prevention team

Motor solution

tbd

Autonomous traffic lights with distance detector

tbd

3D printed STOP + 50 sign

tbd

Our mobile test track

tbd

Our dinos

tbd

Autonomous mode and object detection

tbd

STOP sign and speed 50 sign

tbd

traffic lights detection

tbd

Lane detector

test our lane detector checkout readme

traffic signs or others

• readme - version Lolo / not finished