Category: Traffic
Inspiration: According to the Insurance Institute for Highway Safety and the U.S. Department of Transportation, there were about 30,000 American fatal vehicle crashes in 2014, resulting in average annual costs of over $240 billion. However, even the cheapest collision detection systems are priced at over $2000, as evidenced through the $2100 cost of the collision-detection system for the Volvo S60. Thus, we identified the need to develop a more accessible collision-avoidance system to prevent automotive accidents and enhance driver passenger, and pedestrian safety.
What _ No Collisions _ Does: Our system uses a 2D camera to accurately detect stationary and moving 3D obstacles, including vehicles, in the user’s path. The user’s vehicle automatically reacts to this dangerous situation by changing its own trajectory and steering in the opposite direction of the oncoming automotive, making it especially useful in cases with sleepy or drunk drivers. Costing under $100, No Collisions represents an accessible, inexpensive, efficient, effective, and practical means of preventing potential car accident from occurring.
How We Built _ No Collisions _ : Our collision-avoidance system comprises of a raspberry pi, arduino microcontroller, and laptop with code in python and C++. The raspberry pi uses Haar feature-based cascade classification to detect obstacles that may potentially collide with the user’s car. The arduino microcontroller acts as an interface between the user’s car and the computer vision system. It automatically directs the user’s vehicle to steer away from incoming obstacles, preventing a collision. _ No Collision _ is tested using the _ Beam NG _ soft-body physics simulator.
Tackling Challenges: Our system initially detected a large number of false positives by detecting nonexistent obstacles in the path of the user’s car. To combat this problem, we used filtering techniques to pinpoint the specific obstacles in the user’s trajectory. In addition, the program on the raspberry pi runs on a single core, which resulted in performance disabilities; to mitigate this issue, we increased the detection range, predicted car movements earlier, and performed evasive maneuvers sooner.
Our Main Accomplishments: We created _ No Collisions _, an accessible, inexpensive, efficient, effective, and practical collision-avoidance system to prevent automotive accidents.
1) Accurate detection of 3D obstacles, including stationary and motionary vehicles, using one 2D camera 2) User’s car automatically steers in the opposite direction of incoming vehicles 3) The sensitivity of the system to obstacles can be changed based on the user’s preferences
What We Learned: We learned to come up with an idea, design a project, set and adapt to self-imposed milestones, and create an accurate, functional program in a span of twelve hours. We programmed the raspberry pi for the first time and used new computer vision techniques. This was our first hackathon, and we are thankful for the opportunity to have fun, learn, and enjoy the experience. Because of CU Hacks 2, we look forward to participating in more hackathons!
Future Directions: Once further refined, this adaptive collision-avoidance system may be used as an add-on device to vehicles around the world. It is especially useful in underserved countries with poor transportation infrastructure to significantly enhance the safety of the driver, passengers, and pedestrians. The GPU may be used to further increase performance, and the program itself may be further optimized for utmost efficiency.
Thank you so much, and please try a demo on our laptop!
Log in or sign up for Devpost to join the conversation.