brand banner

DriveSafe AI

Driver risk assessment from dashcam footage, using a ResNet50 + LSTM model to classify driving as safe, collision, tailgating, or weaving.

Updated · 1 min read · #project · #computer-vision · #machine-learning

DriveSafe AI

Driver risk assessment from dashcam footage, using a ResNet50 + LSTM model to classify driving as safe, collision, tailgating, or weaving.

Driver Risk Assessment using Dashcam Footage and Machine Learning

The DriveSafe model uses transfer learning on a pre-trained ResNet50 model to capture image features, and adds an Long Short-Term Memory layer (LSTM) to capture temporal data from video input.

The model takes in dashcam-like video footage and predicts what happens throughout the video according to these 4 categories: safe, collision, tailgating, weaving.


Resources

Official Documentation

Kaggle

  • Download model from Kaggle here.
  • Download dataset from Kaggle here.

GitHub

DriveSafe v4 model training plot

  • DriveSafe: Video Classification Notes (76%) — Early DriveSafe notes on driving behavior classification, training data, model evaluation, and ways to incorporate temporal information from video.
  • AirBoard (31%) — AirBoard is a portable digital whiteboard system engineered to allow teachers to make handwritten annotations on any flat surface using infrared diodes (~940nm), optical sensors, Computer Vision, and perspective transform.
  • The Kernel Trick (SVM) (29%) — Understanding the kernel trick in SVMs, focusing on the roles of alpha and the inference formula.
  • Handwired: Programs as Neural Networks (29%) — How Handwired replaces control flow with numeric data flow, and what that does—and does not—make possible.
  • Bayesian Inference: Alice’s Coin, Likelihood, MLE, and MAP (28%) — One hidden coin connects conditional independence, Bayesian updating, prediction, Naive Bayes, and maximum-likelihood estimation.
  • Data Augmentation for Bias Term (27%) — Adding a constant feature folds the intercept into linear regression’s weight vector without conflating it with statistical bias.
  • yang-ts-parser (26%) — CLI tool to automagically parse and compile YANG models into Typescript types with respect to Nokia SDN extended RESTCONF implementation. Lexer/parser grammar based on ANTLRv4.
  • Maximum Likelihood and Cross-Entropy (26%) — A mathematical textbook connecting likelihood, entropy, KL divergence, statistical estimation, and probabilistic machine learning, with worked examples and solved exercises.

© 2026 James Yap

Personal Website and Knowledge Base