Tracking and Pedestrain Paper Reads

Paper #1 - Object Tracking for Autonomous Driving Systems
keynotes

• The Different components of Autonomous driving System

I'm keen on Tracking - Prediction - Planning aspects
MOT - Multiple Object Tracking

• Record and track moving objects over time
• In Autonomous System tracking cars, pedestrians, signals, passing vehicles everything is key

Deep Learning in Video Multi-Object Tracking

• Detection - Object Detection, Bounding boxes
• Feature Extraction - CNN or Siamese networks employed
• Affinity - Cosine Similarity, Euclidean distance, Intersection over union
• Association - Matching trackets

Real-world challenges

• Camera Motion
• Occlusions

Deep Sort - DeepSORT extends SORT by incorporating appearance information for each tracked object into the affinity stage. The underlying Kalman filter and target state definition are unchanged, but a convolutional feature extraction model is used to process bounding box crops of targets into feature vectors

Novelty
Distractor-Aware Siamese Region Proposal - Networks - DaSiamRPN is a single object tracker that combines a Siamese feature extraction network with a novel training strategy to learn distractor-aware features.

Paper #2  - Estimating Pedestrian Densities, Wait Times, and Flows with Wi-Fi and Bluetooth Sensors

Key Notes
Tracking Approaches

• Infrared, Thermal Sensors
• Bluetooth, Mobile Phones
• Automated Pedestrian Counting Technologies

Pedestrian Counting Techniques

• Anonymize, filter and aggregate traces of pedestrians
• Wifi, MAC-based approach

Paper #3 - Autonomous Vehicles that Interact with Pedestrians: A Survey of Theory and Practice
Key Notes
Different pedestrian vs Environment factors are

Decision-making Factors

SIMPLE ONLINE AND REALTIME TRACKING WITH A DEEP ASSOCIATION METRIC
Key Notes

• Simple online and realtime tracking (SORT)
• Standard Kalman filter with constant velocity motion and linear observation model
• Kalman filtering, also known as linear quadratic estimation (LQE)
• By using simple nearest neighbor queries without additional metric learning

SORT for Tracking - Git Project

Good Reads
ML6 Internship: Pedestrian tracking over multiple non-overlapping camera viewpoints

Keep Thinking!!!

Day #280 - Human detection and Tracking

Project #1 - Human Detection and Tracking 

Overview

• Detecting a human and its face in a given video and storing Local Binary Pattern Histogram
• Recognize them in any other videos
• Local Binary Pattern Histogram - type of visual descriptor, clockwise direction check neighbour values

Execution Steps
Clone the project
Step 1 - python create_face_model.py -i data
Step 2 - python main.py -v video

Project #2 - Person-Detection-and-Tracking  (Pending Execution)

Overview

• The person detection in Real-time is done with the help of Single Shot MultiBox Detector
• Single Shot MultiBox Detector
• Tracking - Kalman Filter is fed with the velocity, position and direction of the person which helps it to predict the future location 

Single Shot MultiBox Detector

• The core of SSD is predicting category scores and box offsets for a fixed set of default bounding boxes using small convolutional filters applied to feature maps
• The key difference between training SSD and training a typical detector that uses region proposals, is that ground truth information needs to be assigned to specific outputs in the fixed set of detector outputs

Execution Steps
Clone the project https://github.com/ambakick/Person-Detection-and-Tracking

Execute - camera.py in Spyder

Project #3 - Tracker Types Demo Project (Pending Execution)

Overview

• Track Multiple faces
• Download and Experiment

Run Below Demos
demo - track multiple faces.py
Multiple_Trackers.py
face_eye.py
distance_to_camera.py

Datasets - Link

Object Motion Detection and Tracking for Video Surveillance
Measuring size and distance with OpenCV
Calculate X, Y, Z Real World Coordinates from Image Coordinates using OpenCV

Happy Learning!!!

Day #279- Multi-Object Tracking

Project #1 - vehicle-speed-check

Clone Repository - Link

On Anaconda prompt,
cd vehicle-speed-check
pip install -r requirements.txt
python speed_check.py

Comments - Very good project to get started. The logic of speed computation with respect to frames per second, pixel movement can be reused in other use cases. Use of dlib correlation tracker. The tracking logic can be reused in other similar implementation

Project #2 - Simple Example code (ROI Based)

Project #3 - Another interesting project from Adrian blog 

Cloned the project and executed the demo. This code does not work in windows 10 though. Someone has fixed the code. The working code is in link 

python multi_object_tracking_fast.py --prototxt E:\Code_Repo\multiobject-tracking-dlib\mobilenet_ssd\MobileNetSSD_deploy.prototxt  --model E:\Code_Repo\multiobject-tracking-dlib\mobilenet_ssd\MobileNetSSD_deploy.caffemodel --video E:\Code_Repo\multiobject-tracking-dlib\race.mp4 --output E:\Code_Repo\multiobject-tracking-dlib\race_output_fast.avi

More Reads
Speed Detection

Multiple object tracking using dlib library, Python

Face Tracking in Python using Xailient Face Detector and dlib, Link

Object Tracker - Link

Happy Learning!!!

Day #278 - Object Tracking - TensorFlow Object Counting API

I came across this project. Fantastic work!! The Detection part needs to be finetuned for the Indian scenario, the tracking seems fine performing decently. You can spot a few false positives, Trucks on the other side of the lane are not detected, Indian Trucks are not well recognized. This can be handled by a custom detection model. Overall the tracking and counter approach can be reused in multiple scenarios.

Clone the project - object_counting_api

vehicle_counting.py - Executed this for some of my highway videos.

The objects passing through the line will be counted and incremented. Minor changes to roi

Output of the same



Happy Learning!!!

Day #277 - Tracking Objects - Deep SORT

What is Deep Sort ?
Simple Online and Realtime Tracking with a Deep Association Metric

How it works ?
It performs Kalman filtering in image space and frame-by-frame data association using the Hungarian method with an association metric that measures bounding box overlap

Paper - Link


Happy Learning!!!

Day #236 - Papers on Person Re-Identification

Paper #1 - Camera Style Adaptation for Person Re-identification

Key Lessons

• Person Reidentification - Given Query Person, Retrieve person from multiple sources
• Challenges - Resolution, Environment, Illumination
• Camera Style Adaptation Approach - unsupervised, camera-invariant property

Techniques

• Input image pairs are partitioned into three overlapping horizontal parts respectively, and through a siamese CNN model to learn the similarity of them using cosine distance

Paper #2 - SIMPLE ONLINE AND REALTIME TRACKING WITH A DEEP ASSOCIATION METRIC
Techniques

• Kalman filtering in image space and frame by frame
• Kalman filter with constant velocity motion

Paper #3 - In Defense of the Triplet Loss for Person Re-Identification
Techniques

• A plain CNN with a triplet loss 

Triplet Loss
Key Lessons

• Look at Anchor, Distance with Positive Example, Distance with Negative Example
• 3 Images at a time Anchor, Positive, Negative Image
• APNN
• d(A,P) = 0.5 Set Margin to achieve it for positive / negative
• L(A,P,N) = Max(||f(A)-f(P)||^2 - ||f(A)-f(N)||^2 + Alpha)
• Chosing Triplets Randomly
• Map Training Set into Triple

Example - Link1, Link2



Happy Mastering DL!!!