March 8

Our prototype and demo are coming together and we are excited to be able to share our progress.

Mechanical Update

The final system assembly was complete for the mechanical and electrical integrated systems. The picture below shows the bike outfitted with the Falcon safety system as we took it out for ride testing.

Mobirise

Outfitted Bike with Prototype

Some problems were encountered during the mechanical assembly process and had to be resolved. The most major problem occurred with faulty ribbon cable extenders connecting the front stereo camera pair to the hardware sync module located in the center of the bike. This failure was suspected to be due to the long distance that the high frequency signals had to transmit over. Since there wasn’t time to order and test new cables, the mechanical system and wiring had to be re-designed to position the hardware sync PCB in the front module and the ribbon cable extensions were removed. The velcro style mount for the rear module was also too loose to securely hold the cameras and would interfere with the users legs while biking. The rear mount was extended and redesigned to be a strong clamping mount.

Assembly of the system highlights some exceptional features designed into the mechanical system. First, the modular component design and robust Skeleton-modeling done allowed for changes to be designed and constructed quickly on the system. The custom machined aluminum heatsinks effectively conducted heat from the computing chips to the external aluminum plates as observed by the warmth of the plate during processing. Lastly, the internal design of the main module allowed for convenient wire routing and convenient access to connectors on the module.

The final mechanical goals for the system is to do exterior touch ups for the symposium debut, specifically sanding the aluminum enclosure and frosting the side/rear LED panels. 

Electrical Update

The assembly and verification of all four PCBAs has been completed. Additionally, the mechanical integration and wiring is finished. This involved some setbacks, namely that a pair of camera FPC cable extensions used in the front module did not work. We hypothesize this is because the signals carried by these cables are high frequency and susceptible to signal integrity factors such as cable impedance, length, and junctions. With some mechanical redesign, a solution was found using a custom FPC extension in the front module, which was more robust.

Another setback was an MCU failure on the main Falcon PCBA. This was likely caused by a short between power sourcing and sinking MCU pins or power rails, or a short to a higher voltage rail. This could have occurred during wiring or from any debris that made its way onto the PCBA. The connections to the board are generally protected from this but on-board connections may not be. The board was reworked to replace the MCU and was further protected physically by covering as much area of the PCBA with Kapton tape as possible.

Basic firmware has been written to test the functionality of the system but power management and Raspberry Pi <-> Main PCBA communications over SPI need to be implemented. This will allow the Pi to control lighting and alerts.

Assembly + Testing Station

Integrated into Main Module

Mobirise

Wiring Work

Software Update

This week, the mechanical and electrical teams finished assembly of the final prototype. The mechanical design did not provide any access to the Raspberry Pi modules, so SSH and VNC were configured to provide remote shell and GUI access. Thorough testing of the cameras, communication interfaces and remaining hardware was performed to verify that all components were functioning normally after assembly.

In our last blog post, we outlined our desire to use 2 Intel Neural Compute Stick 2 devices to accelerate object detection. By coordinating with mechanical, we were able to install a second Neural Compute Stick 2 in the rear module. The configuration is now as follows:
        • The rear module contains the rear stereo cameras and Neural Compute Stick 2. It performs object detection for rear obstacles and their positions are sent to the main module Raspberry Pi via Ethernet.
        • The main module contains the front stereo cameras and a single Neural Compute Stick 2. This hardware performs object detection for front and side obstacles.

Finally, our team began collecting sample footage from all cameras in multiple scenarios such as the brake check and right hook to begin testing our object detection and risk models. We will be tuning these models offline and adapting the software to run in real-time once we are satisfied with the results.

App Development Update

With the app navigation and user authentication complete, software efforts have shifted to the visualization of the Falcon Safety system. Over the past week, the team has been developing the mobile animation of the surroundings and the backend of the application to save the recorded trips on the cloud for later viewing. In the process, the team experienced some problems with the overflow of the objects in the visualization. For example, the following screen capture illustrates the overflow of vehicles below the visualization bounding box.

Mobirise

Screen Navigation Layout

After a lengthy debugging process, it was discovered that the root cause of the problem was in the styling of the wrapper component that was displaying the visualization component above. The resolved visualization is illustrated in the figure (left) below. After solving this problem, we were able to add additional features to the visualization. To demonstrate the forward motion of the cyclist, the background (road) is illustrated to be moving backwards. The following figures (middle and right) illustrate this motion in two subsequent frames.

Mobirise

Resolved Visualization

Mobirise

Simulated Motion - Frame 1

Mobirise

Simulated Motion - Frame 2

So build away, we do.

Come back and check out our blog for regular updates.

University of Waterloo
  • Group 9
  • Falcon Safety