CS 585, Spring 2024 PA3: Tracking Vehicles on the Commonwealth Avenue Code and data: https://drive.google.com/drive/folders/16Eml-MCgg-_giw53DARjS9ury5HuOpU-?usp=sharing Summary of the Assignment This assignment has two parts: 1. Single Object Tracking with a

Bayesian Recursive Filter: Given detected 2D locations of a vehicle over multiple frames in the video, you need to implement an alpha-beta filter or a Kalman filter to generate a smooth 2D track of the vehicle based on these 2D observations. 2. Multi-Object Tracking and Data Association: Given the bounding boxes of multiple objects detected in the video, you need to track them by assigning a unique ID to each object over the video as long as they are detected. THE MOST IMPORTANT RULES: 1. No deep learning for detecting and tracking! 2. No video forgery! Once we find evidences indicating the submission violates the above rules, your final score of PA3 is 0, no matter how much you implemented that do not violate them. Part 1: You are given a list of centers for a vehicle over frames in a video, formatted as [[x_1,y_1],..., [x_n,y_n]. The object location might be missing in certain frames, annotated as [-1,-1]. Your task is to build a tracker that estimates the 2D trajectory of the vehicle and smoothes the 2D track with an alpha-beta filter or Kalman filter. {"obj": [[-1, -1], [312, 228],..., [166, 100], [166, 101] Red: track based on observed centers Blue: ideal track smoothed (e.g. by filter) Grading Criteria 1. Code (10 points + 10 extra credits) a. b. We are going to check whether your code includes deep learning code and if it can generate proper tracks in a video. (Easy 10 points if you are honest in what you've done.) Optional: Implement a Kalman filter (instead of the Alpha-Beta filter) to estimate the vehicle track. (10 extra pts) 2. Tracks (20 points) a. Write your smoothed tracks in the format that is the same as the given one. However, your track should not include any [-1,-1]. We will compare the track with the one based on the observed centers. If they are similar enough, you will get full credit. If not, points will be deducted proportionally according to how dissimilar they are. b. Check part_1_object_tracking.json as an submission example. 3. Visualization (20 points) a. A.gif or .mp4 that draws the track of the object. b. Your track should be following the moving vehicle (10 pts), and it should be sufficiently smooth (10 pts)./nInstructions Complete the computer vision assignment as instructed in CS585_PA3.pdf. You are expected to complete in Python/Jupyter Notebook. Starter code and data is provided in CS585_PA3-starter_codes.zip. The submission file should include the code files(.py or .ipynb files) and the demo videos as well as exported solution videos. The other two files provide can work as references.