Measuring Exit Velocity and Launch Angle Without Radar: A Computer Vision Approach

In the modern baseball landscape, two metrics reign supreme: Exit Velocity (EV) and Launch Angle (LA). Together, they define the quality of contact and predict the probability of a hit. Traditionally, measuring these metrics required expensive radar systems like TrackMan or Rapsodo—tools costing thousands of dollars, making them inaccessible to the average amateur player, youth coach, or local league.

However, the rise of edge-compute AI and high-frame-rate mobile cameras is changing the game. By applying computer vision algorithms to standard smartphone videos, we can now calculate exit velocity and launch angle with remarkable accuracy without using radar hardware. Here is how it works under the hood.

The Physics of Ball Tracking in Video To calculate exit velocity and launch angle from a video file, the system must perform three distinct computational phases: object detection, spatial calibration, and kinematic calculation.

#### 1. Object Detection and Instance Segmentation
First, the video frames are passed through a lightweight convolutional neural network (CNN) trained specifically to detect baseballs in flight. Because a hit baseball travels at speeds exceeding 90 mph (132 feet per second), it often appears as a motion blur rather than a perfect sphere on standard 60 fps or 120 fps video.

The detection model is optimized to identify the "blur tail" and estimate the true centroid of the ball at each frame. By linking these centroids across consecutive frames, the system constructs the ball's flight path.

#### 2. Spatial Calibration (The Coordinate Problem)
A video is merely a 2D grid of pixels. To calculate real-world physical values like miles per hour, the system must map pixel coordinates $(x, y)$ to real-world physical dimensions (meters or feet). This is achieved through spatial calibration.

Our algorithms calibrate the space using known physical reference points in the frame. The most common references include:
* The Home Plate: Standard home plate width is 17 inches.
* The Batter's Box: Standard lines are 6 feet by 4 feet.
* The Batter's Height: Using MediaPipe skeletal tracking, the system estimates the player's height and uses it as a scaling factor.

By calculating a pixels-per-foot ratio, the system can translate the ball's pixel displacement between frames into physical distance.

#### 3. Kinematic Calculation
Once the distance is calibrated, the system measures the ball's displacement immediately after contact. Let's look at the math:

$$\text{Velocity} = \frac{\Delta \text{Distance}}{\Delta \text{Time}}$$

Where $\Delta \text{Time}$ is defined by the frame rate of the video (e.g., $1/60$ seconds or $16.67\text{ ms}$ per frame for $60\text{ fps}$ video).

By tracking the ball's centroid for the first 3 to 5 frames after the bat-ball collision, the system calculates the launch velocity vector. The launch angle is computed by finding the angle of the velocity vector relative to the horizontal home-plate plane:

$$\theta = \arctan\left(\frac{y_2 - y_1}{x_2 - x_1}\right)$$

The Importance of the Sweet Spot Optimizing exit velocity and launch angle is the key to offensive production. Data shows that balls hit with an exit velocity of 95 mph or higher are classified as Hard Hit balls. At this speed, the batting average increases exponentially.

However, exit velocity alone is not enough; it must be paired with the correct launch angle:
* Ground Balls: Under 10 degrees.
* Line Drives: 10 to 25 degrees (the highest batting average zone).
* Fly Balls: 25 to 50 degrees (the home run zone when paired with 95+ mph EV).
* Pop-ups: Over 50 degrees.

The combination of an exit velocity above 98 mph and a launch angle between 8 and 32 degrees is called a Barrel. Hitters who consistently find this "barrel zone" dominate the offensive leaderboards.

Challenges and Future Horizons While computer vision has democratized data collection, it is not without challenges. Camera angle (perspective distortion), lighting conditions, and frame rates can impact accuracy. For instance, a camera positioned directly behind the plate will measure launch angle easily but struggle with exit velocity due to foreshortening along the z-axis. To overcome this, the Precision Baseball Analytics platform uses multi-angle perspective correction matrices to adjust for camera positioning.

As mobile processing power continues to grow, we are moving toward a future where real-world, high-fidelity biomechanical feedback is accessible in every local batting cage, helping athletes realize their potential without financial barriers.