HIRI ICPS: Enabling Quay container trucks to park more accurately than seasoned drivers.

System Overview and Technical Background

As the core carrier for container handling and transshipment, the Quay’s docking positioning accuracy and efficiency directly affect the port’s overall operational rhythm, handling safety, and throughput capacity. Traditional guidance schemes primarily rely on manual command and ground/air signage, requiring no complex sensing equipment; however, they suffer from low precision and heavy reliance on operator experience. Drivers must repeatedly adjust their alignment, and commanders are required throughout the process, resulting in high labor costs and significant efficiency losses due to fatigue and experience, which in turn leads to excessively long single-container turnaround times.

Subsequent sensor‑assisted solutions, such as laser scanners, have improved alignment accuracy; however, in the complex outdoor environments of ports, they are susceptible to interference from fog, rain, and dust. When the laser optics become contaminated, detection accuracy deteriorates markedly. Moreover, these systems are costly, have limited service life, and incur high maintenance expenses, making large‑scale deployment challenging.

01 Opportunities in Technological Evolution
With the advancement of computing power and the widespread adoption of big data technologies, deep learning—particularly convolutional neural networks (CNNs)—has achieved breakthroughs in the field of image recognition. Compared with traditional methods, deep learning eliminates the need for manual feature engineering and can adaptively learn data patterns, offering a new approach to addressing complex nonlinear problems in industrial automation scenarios. HIRI FUTURE has kept pace with this technological trend by integrating visual perception with neural networks, launching the Intelligent Chasiss Positioning System (ICPS) to address the shortcomings of conventional solutions in terms of accuracy, efficiency, and level of intelligence.

Definition of the Intelligent Chasiss Positioning System
HIRI FUTURE’s Intelligent Chasiss Positioning System (ICPS) is a vision-based Intelligent Chassis Guidance System that uses high-definition cameras to capture real-time footage of the operating lanes. By leveraging redundant verification between dual cameras and integrating deep-learning models, it rapidly detects containers and chassis frames, locates key components, and guides drivers to precise docking positions via industrial displays and voice prompts.

The ICPS system boasts core advantages:

High robustness: Over 100,000 data samples have been collected and annotated, covering challenging conditions such as varying illumination, occlusions, nighttime scenes, rain, fog, and snow, ensuring stable model performance.

Real-time performance: Single-frame recognition as fast as 40 ms, with support for 24/7 continuous operation.

Multi-vehicle compatibility: Supports a variety of vehicle types, including internal shuttle carriers, external shuttle carriers, unmanned shuttle carriers, IGVs, AGVs, and more.

Advanced features: Supports dual-container seam detection and tall/container height judgment, outputs TTDS protection signals based on Spreader dimensions, and can calculate the Trolley’s target position and the container truck’s target orientation.

Land-side automation operations: detection of various container types

Supports real-time guidance and secondary guidance for yard crane repositioning.

02 Where does centimeter-level accuracy come from?

— Unveiling ICPS Spatial Perception Technology

2.1 Industry Challenges: Why Can’t Ordinary Cameras Achieve Centimeter-Level Precision?

Wide-angle lens distortion: To cover a wide operational roadway, cameras are often equipped with wide-angle lenses, resulting in edge‑image distortion (barrel or pincushion distortion), which can introduce direct measurement errors of several tens of centimeters.

Inherent limitations of monocular ranging: it lacks depth information, and conventional methods rely on ground‑plane assumptions or manual calibration, making them difficult to adapt to varying vehicle types and cargo‑box heights.

Environmental interference—such as changes in lighting, rain and fog, nighttime conditions, and reflections from shipping containers—can cause drift in the localization performance of conventional computer vision algorithms.

To address these issues, we cannot rely solely on stacking more model tiers; instead, a comprehensive spatial perception technology pipeline is required. HIRI FUTURE has broken down this pipeline into three core steps that are amenable to engineering implementation.

2.2 Core Technological Roadmap: From “Seeing” to “Measuring Accurately”

Step 1: Spatial Reference Mapping — Laying Out the “Coordinate Grid”

Wide-angle lenses cause image edges to curve, and direct measurements can result in errors of several tens of centimeters. The system first applies a distortion-correction algorithm to “straighten” the container edges and vehicle frame structure that appear bent in the image—though they should be straight—thereby restoring the true geometric relationships of the real world. It then establishes a mapping from pixel coordinates to a unified virtual plane coordinate system, effectively laying down a “centimeter‑level coordinate grid” on the image, ensuring that all subsequent measurements are based on an accurate physical reference.

Step 2: Semantic Information Extraction — Understanding “Who Is Who”

The system employs deep learning models to detect objects such as container trucks, trailers, containers, corner castings, and guide plates, and extracts their semantic attributes (type, key points, and standard dimension constraints). This effectively instructs the computer: which elements in the image are chassis, which are boxes, and which are corner fittings. By leveraging prior knowledge—such as the standard spacing between corner fittings—as strong geometric constraints, the system achieves a transition from “seeing pixels” to “understanding the scene.”

Step 3: Spatial Information Perception — Calculating “how much to deviate and where to stop”

Building on the “coordinate grid” and “semantic labels,” the system dynamically selects the optimal reference plane—such as the trailer’s plane or the container’s top surface—based on the target category. Using the bogie‑mounted equipment coordinate system as a reference, it applies a homography transformation to convert each target’s pixel coordinates into two‑dimensional virtual‑plane coordinates. The system ultimately outputs the precise position and orientation angle of each target, along with key metrics such as frame‑edge deviation and inter‑box seam width.

In addition to the three core algorithms mentioned above, ICPS incorporates numerous engineering‑level safeguards in its production implementations:

2.3 HIRI Solutions
Binocular Camera Redundancy Verification

By adopting a dual-HD camera configuration, each camera performs independent detection while simultaneously complementing information from both perspectives. Even if one camera is affected by environmental interference, the camera feed remains usable, thereby enhancing the system’s reliable performance.

Camera image correction

By estimating the parameters of the distortion model via a calibration algorithm, we “straighten” line segments that appear curved in the camera image but are straight in the real world, thereby correcting the distortion introduced by wide-angle lenses and ensuring accurate mapping from image coordinates to real-world coordinates.

End-to-end model localization

From image input to precise recognition of the target task, the process encompasses camera stream acquisition, image correction, object detection, and spatial coordinate computation, achieving pixel-level error control.

Spatial Datum Transformation

By fusing the camera calibration results with the object detection results and integrating PLC signal information, we compute the planar coordinates of the crane’s main carriage and Trolley directions, estimate the container truck’s true physical position and its corresponding yaw angle, and, through task matching, calculate the edge position of the container truck’s trailer, the corner positions of the container, and the width of the container seams.

It is precisely this technical approach—“laying out a coordinate grid → affixing labels → calculating positions”—combined with binocular redundancy and an end-to-end Procedure, that enables ICPS to maintain stable centimeter-level positioning accuracy even in the complex environment of a port.

03 Beyond Positioning: HIRI ICPS Practical Capacity Demonstration

Guiding traditional internal container trucks: Drivers can park precisely in one go, thanks to the industrial display and voice prompts—no on-site guidance required.

Truck Guidance: Supports real-time and secondary guidance for external container trucks (external towing), and is compatible with various container types.

Guidance for unmanned straddle carriers/IGVs: Directly outputs the target parking-space pose for closed-loop control by the Autonomous driving system.

Double-container seam detection + TTDS protection: detects the gap width between two containers, identifies high or low containers, and coordinates with the Spreader to prevent collisions.

Trolley target position calculation: Pre-calculate the optimal alignment position of the Spreader Trolley to improve operational efficiency.

04 Make Every Stop More Relaxed

HIRI FUTURE’s Intelligent Chasiss Positioning System (ICPS) replaces traditional laser‑based and manual solutions with “visual perception + spatial computing,” achieving centimeter‑level positioning using minimal hardware while offering advantages such as low deployment costs, high environmental robustness, and compatibility with multiple vehicle types. As the trend toward intelligent and unmanned Quays gains momentum, the system will continue to evolve: expanding its dataset to include extreme weather conditions, integrating multi‑sensor fusion including millimeter‑wave radar, and extending its applications from chasis guidance to automated container stacking in the Stack yard and fully autonomous horizontal transport and global scheduling, thereby helping ports achieve end‑to‑end automation of their operations.