Self-driving cars can be defined as Autonomous Vehicles (AV) that do not require of human intervention for navigational purposes. Perception systems allow the vehicle to be aware of its surroundings: obtaining information, evaluating it and taking the optimal action at each time step. Fully-AV are limited to the Real-Time domain, the system does not count with any prior knowledge of its surroundings, increasing the difficulty when elaborating new models (trade-off between power and accuracy). It is necessary to obtain fast results while still keeping an eye on the quality/reliability of this outcome. The state-of-the-art for Real-Time object detection models does not use past information as an input to the current state.
The goal of this assignment is to apply segmentation techniques to reconstruct a map that will help the self-driving car to navigate from point A to point B. The segmentation will include the information of previous time frames in an attempt of improving the robustness in the detection