CleverCane

With the DragonBoard™ 410c we were able to make a deep neural network inference system with TensorFlow. This system uses trained models, computer vision techniques and video streams to detect stop signs, traffic lights, and pedestrian crossing areas. It is aimed at assisting the visually impaired with navigating urban traffic environments.

Materials Required / Parts List / Tools

• DragonBoard 410c
• 12V wall adapter
• Maxbotix MB7077
• Step Up Voltage Regulator QYUS-SCA-01
• Logitech Webcam C615
• Outdoor Tech Kodiak 2.0
• USB-to-TTL Serial Cable 954
• Push Button 2751566
• Signal Inverter BOB-11189
• Speaker

Source Code / Source Examples / Application Executable

• Source Code
• Tensorflow Object Detection API
• Object Detection App framework

Build / Assembly

The sonar and camera are both mounted at the two front facing holes of the plastic chassis. The sonar is mounted at the top with the camera below. Connect the MB7077 sonar to the MAX232 signal inverter and then to the USB-TTL logic converter which then connects to one of the DragonBoard 410c USB hubs. This provides serial streaming of sonar data. The webcam connects directly into the USB hub. Power the system by running the 5-9V step up regulator from the REI power pack to the J1 power connection on the DragonBoard 410c. Connect the toggle switch from the 1.8V GPIO power rail from the DragonBoard 410c (pin 35) to GPIO pin 23. Connect the speaker signal lines to the analog expansion pins 1 and 2 (SPKR_OUT_P and SPKR_OUT_M).

Usage Instructions

To get the system running, you must go into the root of the source code. (/home/linaro/jobs/) and run "python run.py" or "./go.sh".Once the system is running, the Clever Cane is used just like a standard cane. Upon detection, an audio signal is triggered to alert the user of objects. The toggle button on the base of the handle toggles the main loop and pauses the detection for the user.