FINAL YEAR PROJECT — COMPUTER ENGINEERING

AI BASED TRAFFIC MANAGEMENT SYSTEM

Real-Time YOLOv8 Detection • Adaptive Signal Control • Smart City Dashboard

Python Flask YOLOv8 Nano React + TypeScript PostgreSQL OpenCV

02

WHY TRAFFIC SYSTEMS ARE FAILING

⏱

FIXED SIGNAL TIMING

Signals run on preset schedules regardless of actual traffic density or congestion

👁

NO REAL-TIME VISIBILITY

Manual monitoring misses sudden congestion events and peak-hour surges

⚠

ZERO ADAPTIVE RESPONSE

System cannot react to accidents, emergencies, or peak-hour congestion

🗄

INSUFFICIENT DATA

No historical storage means no learning, no optimization, no planning

💡 "Urban traffic congestion costs cities billions in lost productivity annually — yet most signals still run on 1970s logic"

03

WHAT WE BUILT

DETECTION & CONTROL PIPELINE

CAMERA
FEED

→

YOLOv8
DETECTION

→

VEHICLE
TRACKING

→

LANE LOAD
CALC

→

ADAPTIVE
SIGNAL

→

SMART
DASHBOARD

🤖

YOLOv8 NANO

Real-time inference at high FPS
Conf: 0.25 | IoU: 0.45

🚦

4-WAY ADAPTIVE CONTROL

Dynamic green phase durations
15 seconds → 60 seconds

📊

FULL-STACK DASHBOARD

React + Flask + PostgreSQL
Live monitoring & analytics

04

TECHNOLOGY STACK

⚙ BACKEND

Python Flask REST API gateway

YOLOv8 Nano Object detection engine

OpenCV Video capture & frame processing

PostgreSQL Historical data storage

Centroid Tracker Multi-vehicle tracking

🖥 FRONTEND

React 18 + TypeScript Dashboard UI framework

Tailwind CSS + shadcn/ui Component system

Recharts Analytics graphs

React Leaflet Traffic map visualization

Three.js 3D animated background effects

Model: yolov8n.pt | Confidence threshold: 0.25 | IoU: 0.45 | Backend: localhost:5000 | Frontend: localhost:5173

05

4-LAYER ARCHITECTURE

TIER 1 — PRESENTATION LAYER

React + TypeScript Dashboard

Vehicle Detection Signal Control Analytics Alerts Settings

▼

TIER 2 — APPLICATION LAYER

Python Flask REST API

/api/detect /api/signal /api/alerts /api/analytics /api/roi-config

▼

TIER 3 — PROCESSING LAYER (CORE ENGINE)

Frame Acq.→ YOLOv8→ Tracking→ Lane Load→ Signal Phase→ Alert Eval→ DB Store

▼

TIER 4 — DATA LAYER

PostgreSQL — historical_traffic OpenWeatherMap API SMTP Server roi_config.json

06

REAL-TIME DETECTION PIPELINE

LIVE DETECTION FEED — ACTIVE

CAR #12 — 42 km/h

TRUCK #7 — 28 km/h

BUS #3 — 35 km/h

NORTH ROI

SOUTH ROI

CARS: 183TRUCKS: 37BUSES: 84TOTAL: 304

304

VEHICLES DETECTED

183 cars | 37 trucks | 84 buses

54–89%

AI CONFIDENCE RANGE

Per detection, per frame

4

CLASSES DETECTED

Car · Truck · Bus · Bike

            Output: [x1, y1, x2, y2, conf, class_id]

            conf=0.25 | IoU=0.45

            Speed = Δcentroid ÷ Δt × px_ratio × 3.6

07

ADAPTIVE SIGNAL CONTROLLER

ADAPTIVE RUNNING

NORTH

45s

65%

SOUTH

—

40%

EAST

—

20%

WEST

—

15%

NS GREEN → NS YELLOW → EW GREEN → EW YELLOW → ↺

LANE LOAD % GREEN DURATION

< 20% 15 seconds

20 – 40% 30 seconds

40 – 70% 45 seconds

≥ 70% 60 seconds

LANE LOAD FORMULA

load = (count × 0.5) + (density × 30) + ((1 − speed_norm) × 20)

08

VIRTUAL LANE DETECTION (ROI)

ROI CONFIGURATION EDITOR

NORTH (4 pts)

SOUTH (4 pts)

EAST (4 pts)

WEST (4 pts)

Edit ROI Save ROI Reload

01 Operator defines 4-point polygons per lane on live video

02 Coordinates stored as normalized values (0.0–1.0) in roi_config.json

03 Ray-casting algorithm assigns each vehicle centroid to a lane

04 Priority order: NORTH → SOUTH → WEST → EAST (prevents double-counting)

05 Lane loads computed and sent to signal controller every frame

            Algorithm: _point_in_poly() — ray-casting

            Storage: roi_config.json (normalized coords)

09

REAL-TIME ANALYTICS

Traffic Flow Analytics — Last 24 Hours

Historical Comparison — Mon vs Tue

Highway

78%

Arterial

71%

Local

82%

4.8 min

AVG WAIT TIME ESTIMATED

/api/analytics/weekly

/api/analytics/compare

/api/minute-vehicle-count

STORAGE

Every 60s

QUERIES

Hourly + Weekly

10

ALERT & NOTIFICATION SYSTEM

0 HIGH 1 MEDIUM 0 LOW 0 RESOLVED

MEDIUM 02:24:44

Congestion Detected

Moderate traffic detected at intersection

Dashboard alert only — no email dispatched

RESOLVED 01:15:22

Congestion Cleared

Traffic returned to NORMAL level

🔴

HIGH PRIORITY

Congestion = SEVERE → Email dispatched via SMTP instantly

🟡

MEDIUM PRIORITY

Congestion = MODERATE → Dashboard alert only, no email

🟢

AUTO-RESOLVED

Congestion clears → Alert resolves, email tracker reset

⚡ Duplicate email prevention — each alert ID triggers max 1 email per activation cycle

11

DATABASE SCHEMA

TABLE: historical_traffic

COLUMN TYPE CONSTRAINT DESCRIPTION

id SERIAL PRIMARY KEY Auto-increment unique identifier

recorded_at TIMESTAMP NOT NULL, INDEXED Snapshot timestamp (fast queries)

vehicle_count INTEGER DEFAULT 0 Total tracked vehicles

average_speed FLOAT DEFAULT 0.0 Average speed in km/h

congestion_level VARCHAR(50) — NORMAL / MODERATE / SEVERE

WEEKLY TRENDS QUERY

SELECT DATE(recorded_at), SUM(vehicle_count)
FROM historical_traffic
GROUP BY DATE ORDER BY date

HOURLY COMPARISON QUERY

SELECT EXTRACT(HOUR FROM recorded_at),
AVG(average_speed) FROM historical_traffic
WHERE DATE = %s GROUP BY hour

Single flat schema — optimized for fast inserts and aggregation queries | Snapshot every 60 seconds

12

TESTING & VALIDATION — 35 TEST CASES

🔍

10/10

PASS

DETECTION & TRACKING

YOLOv8 reliable in daylight, speed within 10% error

🚦

8/8

PASS

SIGNAL CONTROL

Adaptive timing, min enforcement, pause/resume validated

🔔

7/7

PASS

ALERTS & EMAIL

Correct trigger/resolve, duplicate prevention confirmed

🗄

6/6

PASS

DATABASE

Idempotent schema, 60s inserts, resilient to DB downtime

🌐

10/10

PASS

API ENDPOINTS

All HTTP codes correct, CORS validated, input sanitized

Testing conducted using pre-recorded traffic videos, mock data injection, and live dashboard monitoring

13

WHERE WE GO NEXT

⚠ CURRENT LIMITATIONS

Single intersection only — no multi-camera support yet

Signal control is simulated — no real hardware integration

No user authentication or role-based access control

Emergency vehicle detection removed from current scope

Speed estimation uses simplified pixel-to-meter conversion

Some analytics data simulated for demonstration

🚀 FUTURE ENHANCEMENTS

Custom AI model trained on Indian traffic conditions

Multi-intersection + multi-camera architecture

Deep SORT advanced tracking algorithm

Reinforcement learning signal optimization

Docker + cloud deployment (AWS / GCP)

Mobile app for remote monitoring

V2X (Vehicle-to-Everything) integration

Environmental emission monitoring

14

WHAT WE ACHIEVED

🤖

YOLOv8 DETECTION

Real-time vehicle detection & classification across 4 classes

🚦

ADAPTIVE SIGNALS

4-way adaptive control — 15s to 60s dynamic green phases

📊

FULL-STACK DASHBOARD

React + Flask live monitoring, analytics & visualization

🔔

ALERT SYSTEM

Automated alert generation & SMTP email notifications

🗄

HISTORICAL ANALYTICS

PostgreSQL storage with hourly & weekly trend queries

✅

35/35 TESTS PASSED

100% pass rate across all functional test categories

"A scalable, AI-driven traffic management prototype — ready for real-world smart city deployment."

Stack: Python • Flask • YOLOv8 • OpenCV • React • TypeScript • PostgreSQL